JP3576654B2 - Exposure determination method, figure extraction method, and face area determination method - Google Patents

Description

但し、各記号の意味は次の通りである。
【００９４】
ＬＭ：倍率スロープ係数。ネガフィルムの種類とプリントサイズとで定まる引き伸ばし倍率に応じて予め設定されている。
【００９５】
ＣＳ：カラースロープ係数。ネガフィルムの種類毎に用意されており、アンダ露光用とオーバ露光用とがある。プリントすべき画像コマの平均濃度が標準ネガ濃度値に対してアンダかオーバかを判定してアンダ露光用とオーバ露光用の何れかを選択する。
【００９６】
ＤＮ：標準ネガ濃度値。
Ｄ ：プリントすべき画像コマの濃度値。
【００９７】
ＰＢ：標準カラーペーパに対する補正バランス値。カラーペーパの種類に応じて決定される。
【００９８】
ＬＢ：標準焼付レンズに対する補正バランス値。焼付けに用いるレンズの種類に応じて決定される。
【００９９】
ＭＢ：光源光量の変動やペーパ現像性能の変化に対する補正値（マスタバランス値）。
【０１００】
ＮＢ：ネガフィルムの特性によって定まるネガバランス（カラーバランス）値。
【０１０１】
Ｋ_２ ：カラー補正量。
Ｋ_１ ：以下の式で表される濃度補正量。
【０１０２】
【数１】

【０１０３】
ここで、Ｋａ、Ｋｂは定数であり、ＦＤは顔領域平均濃度である。
また、上記（８）式の濃度補正量Ｋ_１ をフィルム検定装置によって求められた補正値とし、カラー補正量Ｋ_２ を次のように顔領域平均濃度を用いて表してもよい。
【０１０４】
【数２】

【０１０５】
但し、Ｋｃは定数である。
更に、上記（１）式の濃度補正量Ｋ_１ 、カラー補正量Ｋ_２ をフィルム検定装置によって求められた補正量とし、（１）式のプリントコマの平均濃度Ｄ_ｉ を顔領域の平均濃度ＦＤ_ｉ に置き換えて露光量を求めてもよい。
【０１０６】
また、顔領域抽出処理において顔領域と判定された領域の各画素毎の濃度（又は色）に対し前記重み係数に応じて重み付けを行って加重平均値を求め、該加重平均値を用いて露光量Ｅ_ｉ を演算することにより、顔領域と判定された領域に対して付与された重み係数を露光量Ｅ_ｉ に反映させるようにしてもよい。
【０１０７】
なお、上記では原画像の記録媒体としてネガフィルム１２を例に説明したが、ポジフィルム等の他のフィルムや、紙等の各種記録媒体に記録された画像を原画像として用いることが可能である。また、複写材料としてカラーペーパを例に説明したが、紙等の他の材料を適用してもよい。
【０１０８】
また、上記では頭部抽出による顔候補領域設定処理において、人物の頭部に特有の形状パターンとして、人物の頭頂部に対応する凸部及び人物の頭部と顔との境界に対応する凹部を用いていたが、これに限定されるものではない。例えば、頭髪のない頭部、頭髪の濃度の低い頭部を抽出するためには、２値化以外の方法による領域分割、或いは画像のエッジの検出を行い、頭部に特有の形状パターンとして凸部のみを用いて頭部を抽出するようにしてもよい。
【０１０９】
更に、上記では画像を複数の領域に分割する方法の一例として２値化を例に説明したが、これに限定されるものではなく、特開平４−３４６３３２号公報に記載されているようにヒストグラムを用いて分割してもよい。また、画像の各画素毎の濃度値を各々異なる複数の方向（例えば図１４（Ａ）に示すように、画像の左右方向（０°）、上下方向（９０°）、及び左右方向に対して±４５°傾斜した方向）に沿って微分することにより画像のエッジを検出し、上記実施形態で説明したラインに代えて、このエッジに対して周知の細線化処理を行って得られたエッジの中心線を用いて処理を行うようにしてもよい。
【０１１０】
なお、上記各方向に沿った微分は、図１４（Ｂ）に示す４種類の微分フィルタを各々用いることで実現でき、画像のエッジの検出は、微分フィルタ出力の大きさが所定のしきい値よりも大きいときに、着目画素をエッジであるとみなすことができる。また、エッジ判定用のしきい値を段階的に変更しながらエッジ抽出処理を繰り返し実施するようにしてもよい。
【０１１１】
また、上記では顔の内部構造を表す形状パターンとして眼部対を用いた場合を例に説明したが、これに限定されるものではなく、鼻、口、眉毛の対等を用いたり、これらを組み合わせて総合的に顔候補領域を設定するようにしてもよい。
【０１１２】
更に、上記では胴体輪郭を表す形状パターンとして、人物の首から肩、肩から腕部分又は胴体下部にかけて連続する輪郭に相当するラインの対を抽出する例を説明したが、これに限定されるものではなく、人物の首から肩にかけての輪郭に相当するラインの対、及び人物の肩から腕部分にかけての輪郭に相当するラインの対を各々別個に検出するようにしてもよい。
【０１１３】
また、上記では人物の各部に特有の形状パターンに基づく顔候補領域の設定処理及び非人物領域判定処理（図２のフローチャートのステップ１０４〜１１２）を順次実行するようにしていたが、これらの処理は、各々他の処理による処理結果の影響を受けないので、上記各処理を行う処理部を別個に設け、前処理として複数領域への画像の分割又は画像のエッジの検出と、凹部、凸部の抽出と、を行った後に、各処理部で並列に処理するようにしてもよい。これにより、処理時間を大幅に短縮することができる。
【０１１４】
〔第２実施形態〕
次に本発明の第２実施形態について説明する。なお、本第２実施形態は第１実施形態と同一の構成であるので、各部分に同一の符号を付して構成の説明を省略し、以下では第２実施形態の作用について説明する。
【０１１５】
本第２実施形態では、画像上での人物を表す図形（請求項６の抽出対象図形に相当）の状態を複数の状態に予め分類している。なお、人物を表す図形の状態としては、例えば画像上での前記図形の大きさ（又は面積）、画像上での前記図形の向き（画像上において前記図形が表す人物の上下方向が何れの方向を向いているがか）、撮影時の視軸に対する被写体としての人物の角度（前記図形が表す人物が該人物の左右方向に沿って何れの方向を向いているか：この方向に応じて人物の顔の向きが正面、斜め、横と変化する）等が挙げられる。
【０１１６】
そして、例えば人物を表す図形の状態を、画像上での前記図形の大きさＬ又は面積Ｓに応じて分類する場合には、大きさＬが大（Ｌ_１ 〜Ｌ_２ ）又は中（Ｌ_２ 〜Ｌ_３ ）又は小（Ｌ_３ 〜Ｌ_４ ）で分類するか、画像上での面積Ｓが大（Ｓ_１ 〜Ｓ_２ ）又は中（Ｓ_２ 〜Ｓ_３ ）又は小（Ｓ_３ 〜Ｓ_４ ）で分類することができる。また、画像上での前記図形の向きに応じて分類する場合は、例えば前記向きを９０°単位又は４５°単位で分類することができる。また、前記図形が表す人物の撮影時の視軸に対する角度に応じて分類する場合は、前記角度が正面か斜めか横かで分類することができる。更に、上記の人物を表す図形の大きさ、向き、前記図形が表す人物の撮影時の視軸に対する角度を組み合わせて分類するようにしてもよい。これにより、画像上での人物を表す図形の状態が、複数の状態に分類されることになる。
【０１１７】
また本第２実施形態では、人物の頭部の輪郭、顔の輪郭、顔の内部構造、胴体の輪郭等の人物を表す図形を構成する形状パターン（請求項６の部分形状パターンに対応）の各々に対し、各形状パターンを探索するための探索条件を規定する探索パラメータを、前記分類した複数の状態に応じて複数種類定めている。
【０１１８】
例として、画像上での人物を表す図形の状態を前記図形の大きさ又は面積に応じて分類した場合に、前記大きさ又は面積に応じて変更する探索パラメータとしては、画像解像度（又は画像分割の粗密度）、頭部輪郭を表す形状パターンを探索する際に黒領域の輪郭の曲率を演算するためのベクトルの長さ（図８（Ｃ）に示したベクトルＰ１Ｐ０、ベクトルＰ０Ｐ２の長さ）、頭部輪郭や顔輪郭や胴体輪郭として探索すべき輪郭の長さの許容範囲、線対称性類似度の演算時に演算対象とする候補対間の距離の許容範囲等が挙げられる。
【０１１９】
画像解像度については、先にも述べたように、人物を表す図形の大きさ又は面積が小さい場合には解像度を高くし、前記大きさ又は面積が大きい場合には解像度を低く（粗く）すればよい。またベクトルの長さについては、前記大きさ又は面積が小さい場合には短くし、前記大きさ又は面積が大きい場合には長くすればよく、また輪郭長の許容範囲及び線対称性類似度の演算対象とする候補対間距離の許容範囲についても、前記大きさ又は面積が小さい場合には短くし、前記大きさ又は面積が大きい場合には長くすればよい。
【０１２０】
また、画像上での人物を表す図形の状態を前記図形の向きに応じて分類した場合に、前記向きに応じて変更する探索パラメータとしては、顔輪郭を表す形状パターンの探索の際の線対象軸の方向（図１０（Ｄ）参照）、顔内部構造としての眼部対を表す形状パターンの探索の際の線対象軸の方向（図１２（Ｂ）参照）、胴体輪郭を表す形状パターンの探索の際の線対象軸の方向（図１３（Ｂ）参照）等が挙げられる。前記向きに応じて上記パラメータを変更することにより、線対称性類似度の演算対象としての候補対の組み合わせ数が絞り込まれる。
【０１２１】
また、画像上での人物を表す図形の状態を前記図形が表す人物の撮影時の視軸に対する角度に応じて分類した場合に、前記角度に応じて変更する探索パラメータとしては、眼部対を表す形状パターンの探索の際の眼部候補領域間の距離（図１２（Ｂ）参照）等が挙げられる。
【０１２２】
次に図１５のフローチャートを参照し、本第２実施形態に係る顔領域抽出処理について、図２のフローチャートと異なる部分のみ説明する。
【０１２３】
本第２実施形態では、ステップ１００で画像データの取込みを行った後に、ステップ１０１では前記分類した複数の状態のうち所定の状態に対応する探索パラメータを取り込む。なお、この所定の状態は、分類した複数の状態のうち、ネガフィルム１２に記録された画像における出現頻度の最も高い状態（例えば、人物を表す図形の大きさ：中、前記図形の向き：人物の上下方向が画像の長手方向と一致する方向、前記図形が表す撮影時の視軸に対する人物の角度：正面）とされている。
【０１２４】
次のステップ１０２〜１１２では、ステップ１０１で取り込んだ探索パラメータに基づいて、第１実施形態と同様にして顔候補領域の設定を行う。前述のように、このステップ１０２〜１１２の処理で用いる探索パラメータは、先に分類した複数の状態のうち所定の状態に対応する探索パラメータであるので、探索対象としての形状パターンの範囲が絞り込まれており、第１実施形態と比較して短時間で処理が終了する。ステップ１１８では、第１実施形態と同様にして人物の顔領域に相当する領域である確度が高い領域の判定を行う。なお、ステップ１１８の処理は、請求項６に記載の、各候補領域毎に整合度を求め、抽出対象図形の特定部分としての人物の顔が存在している確度が高い候補領域を判断することに対応している。
【０１２５】
次のステップ１１９では、上記処理により、人物の顔に相当する領域である確度が非常に高い領域が抽出されたか否か判定する。この判定は、ステップ１１８の顔領域総合判定により演算された総合的な重み係数（請求項６の整合度）が所定値以上の顔候補領域があったか否かに基づいて判断することができる。上記判定が肯定された場合にはステップ１２４へ移行し、抽出された人物の顔に相当する領域である確度が非常に高い領域を顔領域と判定し、ステップ１２６で判定結果を出力して処理を終了する。
【０１２６】
一方、ステップ１１９の判定が否定された場合にはステップ１２１へ移行し、各状態に対応する全ての探索パラメータを用いて処理を行ったか否か判定する。判定が否定された場合にはステップ１２３へ移行し、処理未実行でかつ出現頻度の高い状態に対応する探索パラメータを取込み、ステップ１０２へ戻る。これにより、新たに取り込んだ探索パラメータに従ってステップ１０２以降の処理が繰り返される。
【０１２７】
このように、出現頻度の高い状態に対応する探索パラメータを順に用いて処理を繰り返すので、ステップ１２１の判定が肯定される前にステップ１９１の判定が肯定される、すなわち分類した各状態に対応する全ての探索パラメータが用いられる前に人物の顔に相当する領域である確度が非常に高い領域が抽出される可能性が高く、顔領域抽出処理の平均処理時間が短縮される。そしてステップ１１９又はステップ１２１の判定が肯定されると、第１実施形態と同様にステップ１２４で顔領域総合判定を行い、ステップ１２６で判定結果を出力して処理を終了する。
【０１２８】
なお、上記では撮影時の視軸に対する人物の角度に応じて変更する探索パラメータの一例として眼部候補領域間の距離を用いていたが、予め設定した顔候補領域に対し、眼部対に対応すると推定される領域を探索し、人物の顔としての整合性を判断する場合には、前記角度に応じて変更する探索パラメータとして、顔候補領域内における眼部候補領域の位置又は線対称軸の位置の偏倚度合い、眼部候補領域対間距離の顔候補領域の幅に対する比率等を適用し、前記角度に応じて前記各探索パラメータの許容範囲を変更するようにしてもよい。
【０１２９】
また、上記では複数種類の探索パラメータを順に用いて処理を行っていたが、これに限定されるものではなく、顔領域抽出部４０に、図１５のフローチャートにおけるステップ１０２〜１１８の処理を各々行う処理部を複数設け（図１６参照）、各処理部において、各状態に対応する複数種類の探索パラメータのうちの何れかを各々用いて処理を並列に行わせるようにしてもよい。図１６の例では、人物を表す図形の状態が状態１〜状態ｎのｎ個に分類され、これに応じてｎ個の処理部６０_１ 〜６０_ｎ を設けており、図１５のフローチャートにおけるステップ１２４の顔領域総合判定及びステップ１２６の判定結果出力は顔領域総合判定部６２で行われる。上記のように構成した場合、各処理部６０_１ 〜６０_ｎ による処理が非常に短い時間で完了するので、人物を表す図形の画像上での状態に拘わらず、顔領域抽出処理の処理時間を大幅に短縮することができる。
【０１３０】
更に、上記では抽出対象図形の特定部分として人物の顔に相当する領域を抽出する例を説明したが、本発明はこれに限定されるものではなく、抽出対象図形全体（上記の例では人物に相当する領域全体）を抽出するようにしてもよい。
【０１３１】
また、上記では、抽出対象図形として人物を表す図形を適用していたが、これに限定されるものではなく、例えば抽出対象図形として人工物を表す図形を適用してもよい。写真フィルムに記録される画像中に存在している人工物は、一般に形状等が未知であるが、人工物を表す図形の部分形状は、非人物領域判定処理（図７）でも説明したように、直線や一定曲率の円弧等で構成されていることが多いことが経験的に知られている。このため、これらを部分形状パターンとして用いれば、抽出対象図形としての人工物を表す図形を抽出することができる。そして、写真フィルムに記録された画像を複写材料に焼付ける場合には、例えば人工物を表す図形が存在している確度が高いと判断した領域の重みが小さくなるようにして複写材料への露光量を決定することができる。
【０１３２】
また、本発明を適用して図形を抽出する対象としての画像は、写真フィルムに記録された画像に限定されるものではない。一例として、部品や製品等の大量生産において、生産された部品や製品等が順に搬送されている状況を撮像すると共に、前記搬送されている状況を表す画像を撮像信号から所定のタイミングで抽出し、抽出した画像から、抽出対象図形としての前記部品や製品等を表す図形を抽出することも可能である。この場合、部品や製品等の形状は予め判明しているので、部分形状パターンは容易に定めることができる。また、本発明により抽出した抽出対象図形が存在する領域は、例えば生産した部品や製品等を自動的に検査するために用いることができる。
【０１３３】
以上本発明の実施形態について説明したが、上記実施形態は特許請求の範囲に記載した技術的事項以外に、以下に記載するような技術的事項の実施態様を含んでいる。
【０１３４】
（１）人物の頭部の輪郭を表す形状パターンを探索対象とする場合には、請求項３記載の発明により分割された領域の輪郭、又は請求項４記載の発明により求められたエッジの中心線より凹部と凸部を検出し、検出した１つ以上の凹部又は凸部の各々の特徴量（円曲度、位置、大きさ、方向）に基づいて、前記凹部又は凸部が人物の頭部の輪郭に対応しているか否かを判断することで、前記形状パターンを探索することを特徴とする露光量決定方法。
【０１３５】
（２）人物の顔の輪郭を表す形状パターンを探索対象とする場合には、請求項３記載の発明により分割された領域の輪郭、又は請求項４記載の発明により求められたエッジの中心線より凹部と凸部を検出し、検出した２つ以上の凹部又は凸部の各々の特徴量（円曲度、位置、大きさ、方向）の関係に基づいて、線対称の度合いにより前記２つ以上の凹部又は凸部が人物の顔の輪郭に対応しているか否かを判断することで、前記形状パターンを探索することを特徴とする露光量決定方法。
【０１３６】
（３）人物の顔構造を表す形状パターンのうち眼部対を表す形状パターンを探索対象とする場合には、請求項３記載の発明により分割された領域、又は請求項４記載の発明により求められたエッジの中心線により分割される領域のうち楕円形状の領域を抽出し、抽出された複数の楕円領域の各々の特徴量（形状、濃度）を用いたマッチング処理により所定値以上の類似度の楕円領域の対を検出し、検出した楕円領域の対の線対称の度合いにより前記楕円領域の対が人物の顔の眼部対に対応しているか否かを判断することで、前記形状パターンを探索することを特徴とする露光量決定方法。
【０１３７】
（４）人物の胴体の輪郭を表す形状パターンを探索対象とする場合には、請求項３記載の発明により分割された領域の輪郭、又は請求項４記載の発明により求められたエッジの中心線より凹部と凸部を検出し、検出した２つ以上の凹部又は凸部の各々の特徴量（円曲度、位置、大きさ、方向）の関係に基づいて、人物の首から肩にかけての輪郭、及び肩から腕部分又は胴体下部にかけての輪郭に相当する凹部又は凸部の組を複数検出し、更にその組同士の線対称の度合いにより人物の胴体の輪郭に対応しているかを判断することで、前記形状パターンを探索することを特徴とする露光量決定方法。
【０１３８】
（５）原画像中の人物の顔に相当する確度の低い領域として、請求項３記載の発明により分割された領域、又は請求項４記載の発明により求められたエッジの中心線により分割される領域のうち、領域の輪郭に含まれる直線部分の比率が所定値以上の領域、又は線対称度が所定値以下の領域、又は凹凸数が所定値以上の領域、又は画像外縁との接触率が所定値以上の領域、又は内部の濃度のコントラストが所定値以下の領域、又は内部の濃度が所定のパターンで変化している、或いは所定の変化パターンを繰り返している領域を判定することを特徴とする露光量決定方法。
【０１３９】
【発明の効果】
以上説明したように請求項１記載の発明は、原画像中に存在する人物の所定の部分に特有の形状パターンを探索し、検出した形状パターンの大きさ、向き及び検出した形状パターンとの位置関係に応じて人物の顔に相当すると推定される領域を設定し、設定した領域に重み係数を付与することを、人物の各部に特有の形状パターンの各々を探索対象として各々行い、人物の顔に相当すると推定される領域として設定した複数の領域の各々の範囲、及び複数の領域に各々付与した重み係数に基づいて原画像中の人物の顔に相当する確度が最も高い領域を判断し、判断した領域の色又は濃度の少なくとも一方に基づいて露光量を決定するので、原画像中の人物の顔に相当する領域及びその周辺の領域の色の影響を受けることなく人物の顔に相当する領域のみを抽出し、該領域を適正に焼付けできるように露光量を決定できる、という優れた効果を有する。
【０１４０】
請求項３記載の発明は、原画像を濃度又は色が同一又は近似している複数の画素で構成される複数の領域に分割し、該複数の領域の各々の輪郭を用いて形状パターンを探索するので、人物の各部に特有の形状パターンを効率良く検出できる、という効果を有する。
【０１４１】
請求項４記載の発明は、原画像中に存在するエッジを検出し、検出したエッジの中心線を求め、求めた中心線を用いて形状パターンを探索するので、人物の各部に特有の形状パターンを効率良く検出できる、という効果を有する。
【０１４２】
請求項５記載の発明は、人物の顔に相当すると推定される領域に符号が正の重み係数を付与すると共に、原画像中の人物の顔に相当する確度の低い領域を判定し、判定した領域に符号が負の重み係数を付与するようにしたので、人物の顔に相当する確度が最も高い領域として人物の顔に相当する領域が抽出される確率が更に向上する、という効果を有する。
【０１４３】
請求項６記載の発明は、抽出対象図形を構成する特徴的な複数の部分形状に基づいて複数の部分形状パターンを予め定めておき、部分形状パターンを探索し、検出した部分形状パターンの大きさ、方向、及び抽出対象図形における前記部分形状パターンが表す部分形状の位置に応じて画像中の抽出対象図形の特定部分又は全体が存在すると推定される候補領域を設定することを、複数の部分形状パターンの各々について行い、各候補領域毎に求めた整合度に基づいて抽出対象図形の特定部分又は全体が存在している確度が高い候補領域を判断するようにしたので、画像中に存在する抽出対象図形を、抽出対象図形が存在する領域、及びその周辺の領域の色の影響を受けることなく適正に抽出することができる、という優れた効果を有する。
【０１４４】
請求項７記載の発明は、複数の部分形状パターンの各々に対し探索条件を規定するパラメータとして、画像上での抽出対象図形の状態に応じて複数種類のパラメータを予め定めておき、部分形状パターンを探索し、候補領域を設定することを複数の部分形状パターンの各々について行い、抽出対象図形が存在している確度が高い候補領域を整合度に基づいて判断して抽出する処理を、複数の状態に対応する複数種類のパラメータの各々を用いて行うか、又は、所定の状態に対応するパラメータを用いて前記処理を行った結果、抽出対象図形が存在している確度が高いと判断した候補領域が無かった場合は、所定の状態と異なる状態に対応するパラメータを用いて前記処理を行うことを繰り返し、抽出対象図形が存在している確度が高いと判断した候補領域が有った場合に前記処理を終了するようにしたので、上記効果に加え、処理時間を短縮することができる、という効果を有する。
請求項９記載の発明は、原画像中に存在する人物の所定の部分に特有の形状パターンを探索し、検出した形状パターンの大きさ、向き及び人物の顔と前記所定の部分との位置関係に応じて人物の顔に相当すると推定される領域を設定し、設定した領域に重み係数を付与することを、人物の各部分に特有の複数種類の形状パターンを探索対象として各々行い、人物の顔に相当すると推定される領域として設定した複数の領域の各々の範囲、及び前記複数の領域に各々付与した重み係数に基づいて、原画像中の人物の顔に相当する確度が最も高い領域を判断するので、原画像中の人物の顔に相当する領域及びその周辺の領域の色の影響を受けることなく、人物の顔に相当する領域の抽出精度を向上させることができる、という優れた効果を有する。
【図面の簡単な説明】
【図１】本実施形態に係る写真焼付装置の概略構成図である。
【図２】第１実施形態に係る顔領域抽出部で実行される顔領域抽出処理のメインルーチンを説明するフローチャートである。
【図３】頭部抽出による顔候補領域設定処理を説明するフローチャートである。
【図４】顔輪郭による顔候補領域設定処理を説明するフローチャートである。
【図５】顔内部構造による顔候補領域設定処理を説明するフローチャートである。
【図６】胴体輪郭による顔候補領域設定処理を説明するフローチャートである。
【図７】非人物領域判定処理を説明するフローチャートである。
【図８】頭部抽出による顔候補領域設定処理の詳細として、（Ａ）は原画像、（Ｂ）は２値化により抽出された黒領域の輪郭のトレース、（Ｃ）は輪郭の曲率の演算、（Ｄ）は凹部及び凸部の抽出を各々説明するためのイメージ図である。
【図９】頭部抽出による顔候補領域設定処理の詳細として、（Ａ）は凹部及び凸部の特徴量の演算、（Ｂ）は頭部の判定、（Ｃ）は顔候補領域の設定を各々説明するためのイメージ図である。
【図１０】顔輪郭による顔候補領域設定処理の詳細として、（Ａ）は原画像、（Ｂ）は凹部及び凸部の抽出・分割、（Ｃ）は顔輪郭の候補となり得る対の抽出、（Ｄ）は線対称性類似度の判定を各々説明するためのイメージ図である。
【図１１】顔輪郭による顔候補領域設定処理の詳細として、（Ａ）は線対称性類似度の判定、（Ｂ）は頭部との境界の探索、（Ｃ）は顔候補領域の設定を各々説明するためのイメージ図である。
【図１２】顔構造による顔候補領域設定処理の詳細として、（Ａ）は眼部の候補となり得る黒領域の抽出、（Ｂ）は黒領域対の線対称類似度の判定、（Ｃ）は顔候補領域の設定を各々説明するためのイメージ図である。
【図１３】胴体輪郭による顔候補領域設定処理の詳細として、（Ａ）は前処理で抽出されたライン、（Ｂ）は線対称性類似度の判定、（Ｃ）は顔候補領域の設定を各々説明するためのイメージ図である。
【図１４】（Ａ）はエッジ検出における微分方向の一例を示す概念図、（Ｂ）は各方向への微分を行うための微分フィルタの一例を示す概念図である。
【図１５】第２実施形態に係る顔領域抽出処理を説明するフローチャートである。
【図１６】顔領域抽出処理を並列に実行するための顔領域抽出部の構成の一例を示す概略ブロック図である。
【符号の説明】
１０ 写真焼付装置
１２ ネガフィルム
１８ 色補正フィルタ
２８ カラーペーパ
３２ ＣＣＤイメージセンサ
４０ 顔領域抽出部
４４ 露光量演算部[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention provides a method for determining an exposure amount,Figure extraction methodAnd face area judgment methodIn particular,A face area determining method for determining an area corresponding to the face of a person present in the original image,An exposure determining method for determining an area corresponding to a person's face existing in an original image and determining an exposure so that the determined area is printed in an appropriate color, and an area in which a graphic to be extracted exists from the image To extract a figure.
[0002]
Problems to be solved by the prior art and the invention
The most noticeable part when viewing a portrait is the face of a person.For example, when printing an original image recorded on a film or the like on a copy material such as photographic paper, the color of the face of the person is appropriate. It is necessary to determine the exposure amount so that the image is printed.
[0003]
For this reason, the present applicant divides the film image into a plurality of predetermined fixed regions based on an empirical rule that in a portrait photograph, the probability that the person is located substantially at the center of the image is high. , Weighting each area so that the weight of the area located substantially at the center of the image becomes heavier, obtaining a weighted average value of the densities of the three colors in each area, and determining the exposure amount based on the weighted average value (See Japanese Patent Application Laid-Open No. 63-80242). However, in the above, if the person is actually located in the vicinity of the approximate center of the image, it is possible to obtain an exposure amount at which the person is printed with an appropriate color, but the person greatly deviates from the center of the image. There is a problem that an appropriate amount of exposure cannot be obtained when it is located at the position.
[0004]
In addition, the present applicant divides a color original image into a large number of pixels, decomposes each pixel into three colors of R, G, and B, performs photometry, and obtains a histogram of hue values (and saturation values) based on photometric data. Is obtained, the obtained histogram is divided for each mountain, it is determined to which of the divided mountains each pixel belongs, and each pixel is divided into a group corresponding to the divided mountain, and a plurality of color original images are provided for each group. (So-called clustering), an area corresponding to a person's face among the plurality of areas is estimated, and an exposure amount is determined based on photometric data of the estimated area (JP-A-Hei. 4-346332).
[0005]
Japanese Patent Application Laid-Open No. Hei 6-160993 discloses that in order to improve the accuracy of extracting a region corresponding to a person's face, a region in contact with the outer edge of an image is determined as a background region and removed or extracted. Converting the area into a line figure and determining whether or not the extracted area is an area corresponding to a person's face based on the shape of a nearby area located around the extracted area and the shape of the extracted area Are also described.
[0006]
However, in the above, a flesh color region such as the ground or a tree trunk is present in the original image, and the hue and saturation of the flesh color region are the same as those of the region corresponding to the face of the person in the original image. If they are similar, there is a possibility that this region is erroneously determined as a region corresponding to a human face. Also, if this skin color area is adjacent to the area corresponding to the face of the person, the skin color area cannot be separated from the area corresponding to the face of the person, and the original image may not be divided into areas within an appropriate range. Was. In the above prior art, the processing is performed on the assumption that any of the divided regions is a region corresponding to the face of a person. Therefore, if the original image cannot be divided into regions in an appropriate range, There is a problem that an area corresponding to the face of the person is erroneously determined, and it is not possible to obtain an exposure amount that can properly print the face of the person.
[0007]
In addition, the problem of erroneously determining a region to be extracted when the original image cannot be divided into regions within an appropriate range is not limited to the extraction of a region corresponding to a person's face, but a specific region from an image. This is a common problem when extracting a region where a figure exists. That is, even if the color or the like of the specific graphic to be extracted is known in advance, there is an area where the hue and saturation of the specific graphic are similar in the image, or the area corresponds to the specific graphic. In such a case, for example, there is a problem that an area where a specific graphic exists cannot be properly extracted.
[0008]
The present invention has been made in consideration of the above fact, and extracts only an area corresponding to a person's face without being affected by the color of an area corresponding to the person's face in the original image and the surrounding areas. It is an object of the present invention to provide an exposure amount determining method capable of determining an exposure amount so that the region can be properly printed.
[0009]
Further, according to the present invention, it is possible to obtain a graphic extraction method capable of properly extracting an extraction target graphic existing in an image without being affected by the colors of the area where the extraction target graphic exists and the surrounding area. Is the purpose.
Also, the present invention provides a face area determining method capable of improving the extraction accuracy of an area corresponding to a person's face without being affected by colors of an area corresponding to a person's face in an original image and surrounding areas. The purpose is to obtain
[0010]
[Means for Solving the Problems]
In order to achieve the above object, an exposure amount determining method according to the invention described in claim 1 searches for a shape pattern specific to a predetermined portion of a person present in an original image, and detects the size and orientation of the detected shape pattern. And setting an area estimated to correspond to a person's face according to the positional relationship between the person's face and the predetermined part, and assigning a weighting coefficient to the set area, a plurality of parts specific to each part of the person. Each type of shape pattern is performed as a search target, and based on the respective ranges of the plurality of regions set as the regions estimated to correspond to the face of a person and the weighting factors assigned to the plurality of regions, the original image The area corresponding to the face of the person having the highest accuracy is determined, and the exposure amount to the copy material is determined based on at least one of the color and the density of the determined area.
[0011]
According to a second aspect of the present invention, in the first aspect of the invention, the shape pattern peculiar to each part of the person is a shape pattern representing a contour of a person's head, a shape pattern representing a contour of a person's face, or a person's face. And a shape pattern representing the outline of the body of the person.
[0012]
According to a third aspect of the present invention, in the first aspect of the present invention, the search for the shape pattern is performed by dividing the original image into a plurality of regions each including a plurality of pixels having the same or similar density or color. It is characterized in that the process is performed using the outline of each of the plurality of regions.
[0013]
According to a fourth aspect of the present invention, in the first aspect, the search for the shape pattern is performed by detecting an edge present in the original image, obtaining a center line of the detected edge, and using the obtained center line. It is characterized by:
[0014]
According to a fifth aspect of the present invention, in the first aspect of the invention, a positive weighting coefficient is assigned to a region estimated to correspond to the set face of the person, and the area corresponds to the face of the person in the original image. It is characterized in that a region having a low degree of accuracy is determined, and a negative weighting coefficient is assigned to the determined region.
[0015]
According to the graphic extracting method of the present invention, a plurality of partial shape patterns are determined in advance based on a plurality of characteristic partial shapes constituting an extraction target graphic, and a partial shape pattern existing in an image is searched for. A candidate for which it is presumed that a specific portion or the whole of the extraction target graphic in the image exists according to the size and direction of the detected partial shape pattern and the position of the partial shape represented by the partial shape pattern in the extraction target graphic Setting an area is performed for each of the plurality of partial shape patterns, and for each set candidate area, a specific portion or the entirety of the extraction target graphic exists based on the relationship with the detected each partial shape pattern. The degree of matching as a region is determined, and based on the degree of matching determined for each candidate region, a candidate region having a high probability that a specific portion or the entirety of the extraction target graphic exists is determined. And the accuracy of a particular part or all are present in the extracted target figure to extract candidate regions judged to be high.
[0016]
According to a seventh aspect of the present invention, in the invention of the sixth aspect, the states of the extraction target graphic on the image are classified into a plurality of states in advance, and a partial shape A plurality of types of parameters are determined in advance in accordance with the plurality of classified states as parameters that define search conditions when searching for a pattern, a partial shape pattern present in an image is searched, and the candidate area is set. Processing for each of the plurality of partial shape patterns, obtaining a degree of consistency for each set candidate area, and determining and extracting a candidate area having a high probability that a specific portion or the entirety of the extraction target graphic exists is extracted. Is performed using each of a plurality of types of parameters corresponding to the plurality of states, or the processing is performed using a parameter corresponding to a predetermined state among the plurality of states. As a result, if there is no candidate area determined to have a high probability that the specific portion or the whole of the extraction target graphic exists, the process is performed using parameters corresponding to a state different from the predetermined state. This process is repeated, and when there is a candidate area determined to have a high probability that the specific part or the whole of the extraction target graphic exists, the processing is terminated.
[0017]
In the invention according to claim 8, in the invention according to claim 7, the state of the extraction target graphic on the image includes the size of the extraction target graphic on the image, the orientation of the extraction target graphic on the image, and the extraction. It is characterized by including at least one of the directions of the object represented by the target graphic.
According to a ninth aspect of the present invention, there is provided a face area determination method, wherein a shape pattern specific to a predetermined portion of a person existing in an original image is searched for, and the size, direction, and face of the detected shape pattern and the face of the person are detected. A region that is presumed to correspond to the face of a person is set according to the positional relationship with the portion, and a weighting factor is assigned to the set region, and a plurality of types of shape patterns unique to each portion of the person are searched. Based on the respective ranges of the plurality of regions set as the regions estimated to correspond to the person's face, and the weights assigned to the plurality of regions, respectively, and correspond to the person's face in the original image. Determine the region with the highest accuracy.
[0018]
According to the first aspect of the present invention, a shape pattern specific to each part of the person existing in the original image is searched, and the face of the person is determined according to the size and direction of the detected shape pattern and the positional relationship with the detected shape pattern. Is set to be equivalent to. The shape pattern specific to each part of the person includes, for example, a shape pattern representing the contour of the head of the person, a shape pattern representing the contour of the face of the person, and the inside of the face of the person. A shape pattern representing the structure, a shape pattern representing the contour of the body of the person, or the like can be used.
[0019]
For example, a region on the original image corresponding to the head (hair portion) of a person has, on its contour, a convex portion having a predetermined curvature corresponding to the top of the person, and a human head below the convex portion. And a concave portion having a predetermined curvature corresponding to the boundary between the face and the face. Therefore, for the head of a person, for example, a convex portion having a predetermined curvature and a concave portion having a predetermined curvature can be used as a specific shape pattern representing the contour of the head. Then, when the above-mentioned shape pattern is detected in the original image, an area estimated to correspond to the face of a person can be set as follows, for example.
[0020]
That is, the face of the person exists at a position adjacent to the lower side of the head (the concave side of the shape pattern), and its size is substantially proportional to the size of the head, and the direction of the head is further Match the orientation. In general, the contour of the face of a person is substantially elliptical. Therefore, when a shape pattern representing the contour of the head described above is detected in the original image, the size and the direction according to the size and the direction of the detected shape pattern and the head and the face of the person are detected. , An elliptical area may be set at a position corresponding to the positional relationship with, and the area may be an area estimated to correspond to a human face.
[0021]
As described above, according to the first aspect of the present invention, it is estimated that the shape corresponds to the face of a person according to the size and direction of the shape pattern specific to each part of the person detected from the original image and the positional relationship with the detected shape pattern. Since an area corresponding to a person's face is set, it is possible to appropriately set an area estimated to correspond to a person's face without being affected by the color of an area corresponding to the person's face and surrounding areas. Conversely, even if the original image does not have an area corresponding to the face of a person but includes a skin color area having a contour similar to the face area, it is possible to prevent erroneous determination of the area corresponding to the face of the person. it can.
[0022]
According to the first aspect of the present invention, each of a plurality of types of shape patterns specific to each part of a person is set as a search target, and a region estimated to correspond to a person's face set based on each detected shape pattern is weighted. Coefficients are assigned to each of a plurality of regions set as regions estimated to correspond to a person's face, and a weight coefficient assigned to each of the plurality of regions corresponds to a person's face in the original image. The area with the highest accuracy is determined. As described above, by performing the processing a plurality of times with each of the plurality of shape patterns unique to each part of the person as a search target, even if the predetermined shape pattern is accidentally detected in the non-human area in the original image, it is detected by accident. An area set based on the shape pattern is not determined as an area having the highest accuracy corresponding to a human face. Since the amount of exposure to the copy material is determined based on at least one of the color and the density of the region determined in this manner, the amount of exposure can be determined so that the region corresponding to the face of a person can be properly printed. .
[0023]
In the search for the shape pattern, for example, as described in claim 3, the original image is divided into a plurality of regions composed of a plurality of pixels having the same or similar density or color, and each of the plurality of regions is Can be performed using the contour of When an image is divided into a plurality of regions as described above, a plurality of regions having a high probability that a shape pattern specific to each part of a person is included in a contour or the like, such as a boundary between a person and a background, is obtained. Therefore, a shape pattern specific to each part of the person can be efficiently detected from the outlines of the plurality of regions.
[0024]
The original image is divided into one-dimensional histograms of hue values or two-dimensional histograms of hue values and saturation values based on photometric data of three colors for each pixel of the original image, for example, as disclosed in JP-A-4-346332. A dimensional histogram is obtained, the obtained histogram is divided for each mountain, it is determined to which of the divided mountains each pixel belongs, each pixel is divided into groups corresponding to the divided mountain, and the original image is divided for each group. It can be divided into multiple areas. Alternatively, the original image may be divided into a plurality of regions by setting a pixel having a large density difference or color difference between adjacent pixels among the pixels constituting the original image as a region boundary. Further, the original image may be divided into a plurality of regions by performing binarization, or the binarization may be performed a plurality of times using a plurality of thresholds having different values, and may be further divided. It may be divided into a plurality of areas by the above combination.
[0025]
In the present invention, the area divided as described above is used to detect a plurality of shape patterns specific to each part of the person, and the setting of the area estimated to correspond to the face of the person is set as described above. This is performed according to the size and orientation of the shape pattern detected from the original image and the positional relationship with the detected shape pattern. Therefore, even if a part of the plurality of shape patterns cannot be detected from the outlines of the plurality of regions, it is possible to set the region estimated to correspond to the face of the person, and, as in the related art, to the plurality of divided regions. Does not need to match the area corresponding to the face of the person.
[0026]
Further, the search for the shape pattern may be performed by previously detecting an edge present in the original image, determining a center line of the detected edge, and using the determined center line. Good. The edge can be detected by, for example, performing differentiation or the like on the density or color of each pixel, and the center line of the edge can be obtained by, for example, a known process such as thinning or ridge detection. . Since there is a high probability that an edge present in an image includes a specific shape pattern in each part of the person, it is possible to efficiently detect a shape pattern specific to each part of the person as in the invention of claim 3. Can be.
[0027]
By the way, an image may include an area having a feature that can be determined to be low in the area corresponding to a person. For example, a region where the ratio of the straight line portions included in the contour is equal to or more than a predetermined value has a high probability of being a region representing an artifact. Further, the human body is substantially line-symmetric with respect to a virtual line dividing the left and right of the human body, but a region where the degree of line symmetry is equal to or less than a predetermined value has a low probability of being a region corresponding to a person. Also, it can be determined that the probability that the region having the number of irregularities is equal to or more than the predetermined value is a region corresponding to a person is low. In addition, since a person is generally more likely to be located substantially at the center of an image, it can be determined that a region having a contact ratio with the outer edge of the image that is equal to or greater than a predetermined value is also located at the periphery of the image. Is low.
[0028]
Further, when the contrast of the density in the region is equal to or less than a predetermined value, it is highly likely that the region is a region representing an object having a smooth or less uneven surface, and the probability of being a region corresponding to a human face is low. . Furthermore, even when the density in the area changes in a predetermined pattern or when the density in the area repeats a predetermined change pattern, it can be determined that the probability that the area corresponds to a person is low. Therefore, as described in claim 5, a sign is assigned a positive weighting coefficient to an area estimated to correspond to the set face of the person, and the probability of the area corresponding to the face of the person in the original image is low. If the area is determined and a negative weighting coefficient is assigned to the determined area, the probability that the area corresponding to the person's face is extracted as the area having the highest accuracy corresponding to the person's face is further improved.
[0029]
In the invention according to claim 6, a plurality of partial shape patterns are determined in advance based on a plurality of characteristic partial shapes constituting the extraction target graphic existing in the image. Note that the extraction target graphic is not limited to a figure representing a person, and may be an object other than a person, for example, a figure representing an artificial object, but when a figure representing a person is applied as the extraction target figure, As the partial shape pattern, similarly to claim 2, a shape pattern representing a contour of a person's head, a shape pattern representing a contour of a person's face, a shape pattern representing an internal structure of a person's face, a body of a person Can be used.
[0030]
According to the invention of claim 6, next, a partial shape pattern existing in the image is searched, and the size and direction of the detected partial shape pattern and the position of the partial shape represented by the partial shape pattern in the extraction target graphic are determined. Setting a candidate area in which a specific portion or the entirety of the extraction target graphic in the image is assumed to exist is performed for each of the plurality of partial shape patterns. As a result, similarly to the first aspect, it is presumed that the specific portion or the entirety of the extraction target graphic exists without being affected by the color of the region where the specific portion or the entirety of the extraction target graphic exists and the surrounding area. Can be set appropriately. As a specific part of the extraction target graphic, for example, when a figure representing a person is applied as the extraction target graphic, a part corresponding to the face of the person can be applied.
[0031]
According to the invention of claim 6, next, for each of the set candidate regions, a specific portion of the extraction target graphic is determined based on the relationship (for example, both size, direction, positional relationship, etc.) with each detected partial shape pattern. Alternatively, the degree of consistency as a region where the entirety exists is determined, and based on the degree of consistency determined for each candidate region, a candidate region having a high probability that a specific portion or the entirety of the extraction target graphic exists is determined, and the extraction target is determined. A candidate area determined to have a high probability that a specific part or the whole of the figure exists is extracted.
[0032]
As described above, the matching degree as a region where the specific portion or the whole of the extraction target graphic exists is obtained for each candidate region, and the candidate region having a high probability that the specific portion or the whole of the extraction target graphic exists is determined. Therefore, even if the candidate region set based on the detection result of a certain partial shape pattern is a region where the extraction target graphic does not exist, it is determined that the specific region or the entirety of the extraction target graphic exists. The extraction target graphic existing in the image can be properly extracted without being affected by the colors of the area where the extraction target graphic exists and the surrounding area.
[0033]
Incidentally, the partial shape of the extraction target graphic on the image changes depending on the state of the extraction target graphic on the image (for example, the size of the extraction target graphic on the image, the orientation of the object represented by the extraction target graphic, and the like). For this reason, when the state of the extraction target graphic on the image is indefinite, the search conditions for searching the partial shape pattern are set so that the partial shape of the extraction target graphic in various states is detected as the partial shape pattern. There is a need to. However, when the range of the partial shape to be detected as the partial shape pattern is expanded, a large number of patterns are detected as the partial shape pattern, and the partial shape pattern is erroneously determined as the partial shape pattern even though it is not actually the partial shape of the extraction target graphic. There is a problem that the number of detected patterns also increases rapidly, and the processing time significantly increases.
[0034]
For this reason, in the invention according to claim 7, the state of the extraction target graphic on the image is classified into a plurality of states in advance, and when searching for a partial shape pattern for each of the plurality of partial shape patterns, A plurality of types of parameters are determined in advance according to the plurality of classified states as parameters defining the search condition. The state of the extraction target graphic on the image includes the size of the extraction target graphic on the image, the orientation of the extraction target graphic on the image, and the extraction target graphic as described in claim 8. The orientation of the object to be represented, and the like. In the above description, the search conditions (parameters for defining) are determined for each type of the graphic to be extracted on the image. Therefore, the range of the partial shape to be detected as the partial shape pattern in each search condition is narrowed, If the search is performed using the search condition, the partial shape pattern of the extraction target graphic in a specific state is searched in a very short time.
[0035]
After defining a plurality of types of parameters, for example, a search for a partial shape pattern and setting of a candidate area are performed for each of the plurality of partial shape patterns, a matching degree is obtained for each of the set candidate areas, and a target pattern to be extracted is specified. In the case where a plurality of processing units for performing a process of judging and extracting a candidate region having a high probability of the existence of a part or the whole are provided, in each processing unit, any one of a plurality of types of parameters corresponding to each state is used. By performing the processing in parallel by using, the processing can be performed using each of the plurality of types of parameters as a whole. Thus, regardless of the state of the extraction target graphic on the image, it is possible to extract a candidate region having a high probability that the specific part or the whole of the extraction target graphic exists, in a very short time.
[0036]
In addition, as a result of performing the above-described processing using the parameter corresponding to the predetermined state among the plurality of states, there is no candidate area determined to have a high probability that the specific part or the whole of the extraction target graphic exists. Repeats the above process using parameters corresponding to a state different from the predetermined state, and there is a candidate area determined to have a high probability that the specific part or the whole of the extraction target graphic exists. The above process may be terminated at this time. In this case, the processing corresponding to each state is performed in series, but before performing all the processing corresponding to each state, a candidate area having a high probability that a specific part or the whole of the extraction target graphic exists is determined. The possibility of extraction is high. Therefore, the extraction target graphic can be extracted in a shorter time than when a plurality of types of search condition parameters are not determined according to the state of the extraction target graphic on the image.
Further, in the above, if the search is performed by using the parameter corresponding to the state of high appearance frequency in order according to the appearance frequency of the extraction target graphic on the image, the time until the extraction target graphic is extracted is further reduced. Can be shortened.
[0037]
According to the ninth aspect of the present invention, similarly to the first aspect of the present invention, a shape pattern specific to a predetermined portion of a person existing in the original image is searched for, and the size, direction, and size of the detected shape pattern are determined. Setting an area estimated to correspond to the face of a person according to the positional relationship between the face and the predetermined part, and assigning a weighting factor to the set area, a plurality of types of shapes specific to each part of the person Each of the patterns is performed as a search target, and based on the respective ranges of the plurality of regions set as the regions estimated to correspond to the face of the person, and the weighting factors assigned to the plurality of regions, respectively, Since the area having the highest accuracy corresponding to the face is determined, the extraction accuracy of the area corresponding to the person's face in the original image is not affected by the colors of the area corresponding to the person's face and the surrounding areas. Can be improved That.
[0038]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, an example of an embodiment of the present invention will be described in detail with reference to the drawings.
[0039]
[First Embodiment]
FIG. 1 shows a photographic printing apparatus 10 to which the present invention can be applied. The photographic printing apparatus 10 includes a transport roller 14 that transports the negative film 12. A light source 16, a color correction filter 18 such as a dimming filter, and a diffusion box 20 are sequentially arranged below the transport path of the negative film 12. A distribution prism 22 for distributing the light transmitted through the negative film 12 in two directions is disposed above the transport path of the negative film 12.
[0040]
On one optical path of the light split in two directions by the splitting prism 22, a projection optical system 24, a black shutter 26, and a color paper (printing paper) 28 as a copy material are arranged in order. On the optical path, a projection optical system 30 and a CCD image sensor 32 are sequentially arranged. The CCD image sensor 32 divides the entire image (one frame) recorded on the negative film 12 into a number of pixels (for example, 256 × 256 pixels), and divides each pixel into R (red), G (green), and B (blue). The light is divided into three colors and measured.
[0041]
The signal output terminal of the CCD image sensor 32 includes an amplifier 34 for amplifying a signal output from the CCD image sensor 32, an analog-digital (A / D) converter 36, and a 3 × 3 for correcting the sensitivity of the CCD image sensor 32. The matrix circuits 38 are connected in order. The 3 × 3 matrix circuit 38 is configured by a microprocessing unit (MPU), and calculates a face area extraction unit 40 in which a program for realizing a face area extraction process described later is stored in a ROM in advance, and an average density of the entire image. Each is connected to the average density calculation unit 42. Further, the face area extracting section 40 and the average density calculating section 42 are connected to the exposure amount calculating section 44. The exposure calculation section 44 is connected to the color correction filter 18 via a driver 46 for driving the color correction filter 18.
[0042]
Next, the operation of the first embodiment will be described. The light emitted from the light source 16 passes through the color correction filter 18, the diffusion box 20, and the negative film 12, is distributed by the distribution prism 22, and is incident on the CCD image sensor 32 via the projection optical system 30. At this time, the black shutter 26 is closed. The CCD image sensor 32 divides the entire screen into a large number of pixels by the incident light, separates each pixel into three colors of R, G, and B, performs photometry, and outputs a photometric data signal. The photometric data signal is amplified by an amplifier 34 and converted into a digital signal by an A / D converter 36, and then the sensitivity of the CCD image sensor 32 is further corrected by a 3 × 3 matrix circuit 38, and a face area extracting unit 40 And the image data is input to the average density calculation unit 42.
[0043]
The average density calculator 42 calculates the average density of the entire screen. The face region extraction unit 40 estimates a portion corresponding to a person's face in one screen as described later in detail, and outputs R, G, and B colorimetric data of the portion. The exposure calculator 44 calculates an exposure using the photometric data of the three colors output from the person area calculator 40 and the average density output from the average density calculator 42, and calculates a negative film using the calculated exposure. The color correction filter 18 is moved via the driver 46 so that the image 12 is printed on the color paper 28, and the black shutter 26 is opened and closed to print the image.
[0044]
The average density of the entire screen calculated by the average density calculation unit 42 is not essential information in the exposure calculation by the exposure calculation unit 44, and the average density calculation unit 42 is omitted. The exposure amount may be calculated using only the photometric data of the three colors output from the region extracting unit 40.
[0045]
Next, with reference to the flowchart of FIG. 2, the face area extraction processing executed by the CPU (not shown) of the face area extraction unit 40 will be described. In step 100, image data is taken in from the 3 × 3 matrix circuit 38.
[0046]
In step 102, the image data is binarized with a predetermined threshold as an example of a method of dividing the original image into a plurality of regions. Hereinafter, an example of a positive image will be described. By this binarization, the original image is divided into a region composed of pixels having a density higher than the threshold (hereinafter, this region is referred to as a “black region”) and a pixel having a density lower than the threshold. The area is divided into a configured area (hereinafter, this area is referred to as a “white area”). As a result, for example, in the original image shown in FIG. 8A, a large number of black regions including the black region 50 corresponding to the head of a person in the original image are extracted as shown in FIG. 8B. become. Further, in step 102, numbering is performed on each of the plurality of divided black regions and white regions in order to identify each of the regions.
[0047]
In the next step 104, face candidate area setting processing by head extraction is performed. This processing will be described with reference to the flowchart in FIG. In step 172, one of the plurality of black regions extracted by the above-described binarization is extracted, and in step 174, the contour of the extracted black region is traced, and the curvature of the outline is calculated. For example, if a large number of black areas are extracted by binarizing the original image shown in FIG. 8A, and a black area 50 as shown in FIG. This can be done clockwise as shown in FIG. In the calculation of the curvature, as shown in FIG. 8 (C), vectors having a predetermined length in a direction substantially along the trace direction and whose start point and end point are in contact with the contour of the black area are sequentially set (FIG. In (C), θ as the curvature can be calculated by the inner product of the vectors P1P0 and P0P2) and the adjacent vectors. Further, the direction of θ (positive or negative sign) can be obtained from the cross product of vectors.
[0048]
At this time, the angle calculation formula is set so that the contour of the black area has a negative value when the contour of the black area is curved rightward and a positive value when the contour of the black area is curved leftward with respect to the traveling direction of the trace. . For example, FIG. 8C shows a portion of the outline of the black region that curves to the right with respect to the traveling direction of the trace, and θ has a negative value.
[0049]
Note that the length of the vector may be fixed, or may be changed according to the perimeter of the outline of the black area. Further, the curvature may be calculated a plurality of times, and the length may be changed stepwise each time.
[0050]
In the next step 176, based on the curvature of the outline of the black region calculated above, a shape pattern representing the outline of the head of the person (the partial shape pattern according to claim 6, where the extraction target graphic is As a figure representing a person, a black region having a concave portion estimated to correspond to the boundary between the head and the face of the person and a convex portion estimated to correspond to the top of the person is extracted in the outline. . The concave portion and the convex portion can be determined based on the trace direction and the direction of θ (positive or negative sign) obtained earlier. That is, the portion where the sign of θ is inverted is the boundary between the concave portion and the convex portion, and when traced clockwise, the sign of θ is positive, and if negative, it can be determined that the portion is a convex portion. . Thereby, as an example, as shown in FIG. 8D, a concave portion and a convex portion in the outline of the black region can be extracted.
[0051]
In step 178, the feature amount of each of the concave and convex portions extracted as described above is calculated. In the present embodiment, the curvature calculated as follows is used as the feature amount of the concave portion and the convex portion. That is, as shown in FIG. 9A, the length of a curve forming the concave or convex portion is L, the distance between both end points Q1 and Q2 of the concave or convex portion is M, and a straight line connecting both end points Q1 and Q2. The height of the concave portion or the convex portion with respect to is set to h, and L ÷ M and h ÷ M are calculated as the degree of curvature. Further, a direction vector V is obtained as a vector heading in the direction in which the concave portion and the convex portion open.
[0052]
In step 180, it is determined whether or not the black area can be determined to be the head (hair portion) of the person, that is, whether or not the probability of the head of the person is high, using the feature amounts calculated as described above. In this determination, the concave portion and the convex portion each having a curvature within a predetermined range are respectively extracted from the outline of the black region, and the ratio of the length of the concave portion and the convex portion to the peripheral length of the outline of the black region is within the predetermined range. And when the consistency evaluation is higher as the head of the person than the position and the direction, the result is affirmative. This consistency evaluation can be performed as follows as an example. First, the consistency in the unit of the concave portion and the convex portion is evaluated by the following procedure.
[0053]
{Circle around (1)} Select any three points on the curve of the concave or convex part (for example, both end points and the center point of the curve, but three points not on the same straight line), and assume a circle passing through each of the selected three points. , The center point of the circle is determined.
[0054]
{Circle around (2)} The distances between all the pixels constituting the curved line of the concave portion or the convex portion and the center point of the circle are obtained, and a curve having a variation in distance of not more than a predetermined value is evaluated as having consistency. If the variation is larger than a predetermined value, the curve is excluded from the evaluation target, or the curve is further divided at a portion where the curvature is large, and the process is performed again.
[0055]
Next, the consistency including the concave portions and the convex portions is evaluated as follows.
{Circle around (1)} Using the center of gravity of the center point of the circle obtained for each curve of the concave portion and the convex portion, a total center point and a total center area (of a radius determined according to the average of the length of each curve) A circular area (an area indicated by a broken line in FIG. 9B) is set. When the center point obtained for each curve falls within the overall center area, or when the center axis of each curve (direction vector V or an extension thereof) converges within the center area, It is evaluated that the consistency as the head of the person is high.
[0056]
Further, instead of the evaluation of the above (1), in the same manner as in the above (2), the dispersion of the distances between all the pixels constituting the curves of the concave and convex portions and the center point of the circle is a predetermined value. In the following cases, it is evaluated that the consistency as the head of the person is high.
[0057]
If the determination in step 180 is affirmative, most of the persons have a concave hair shape at the boundary between the human hair and the face. An elliptical area inscribed in the concave portion is defined as a face candidate area (an area estimated to correspond to the face of a person, a specific portion of the extraction target graphic (person) Is set as a candidate area) where it is estimated that the face exists. In step 184, a positive weighting coefficient is assigned to the face candidate area set above. This weighting coefficient can be set to a value such that the higher the evaluation, the heavier the weight, in accordance with the evaluation result of the above-described black region as the head of the person.
[0058]
After executing step 184, the process moves to step 186. If the determination in step 180 is negative, the process proceeds to step 186 without performing any processing. In step 186, it is determined whether or not the above processing has been performed on all the black regions extracted by the binarization. If the determination in step 186 is negative, the process returns to step 172, and steps 172 to 186 are repeated. If the determination in step 186 is affirmative, the face candidate area setting processing by head extraction is terminated, and the routine goes to step 106 in the flowchart of FIG.
[0059]
In step 106, a face candidate area setting process based on the face contour is performed. This processing will be described with reference to the flowchart in FIG. In step 189, for each of the plurality of black regions and white regions obtained by dividing the original image by binarization in step 102 of the flowchart of FIG. 2, steps 174 and 174 of the flowchart of FIG. In the same manner as in 176, the contour of each area is traced, and the curvature of the contour is calculated to extract the concave and convex portions. Thereby, as an example, a concave portion and a convex portion as shown in FIG. 10B are extracted from the original image shown in FIG.
[0060]
After extracting the concave portions and the convex portions, attention is paid to the longer one of the curves extracted as the concave portions or the convex portions so that the curvature (angle θ) falls within a predetermined range or the circular curvature. May be further divided so that is less than or equal to a predetermined value, or by roughly classifying the angle in the normal direction of the curve. For example, when the curve (2) and the curve (3) are extracted as a continuous curve in FIG. 10 (B), they are divided into two curves because of their large curvature. In FIG. 10B, the curve (1) is also an object of division. In the following, curves obtained by dividing the outline of each region extracted as described above are collectively referred to as “lines”.
[0061]
In step 190, a pair of lines estimated to correspond to the sides of the face of the person is extracted as a shape pattern (partial shape pattern according to claim 6) representing the contour of the face of the person. Specifically, among the lines extracted above, the respective directional vectors V face each other, the magnitude of the intersection angle of the respective directional vectors V (or an extension thereof) is within a predetermined value, and All pairs of lines whose similarity between the length and the degree of curvature are equal to or larger than a predetermined value are extracted as pairs of lines that can be candidates for the face contour. For example, when lines (1) to (7) shown in FIG. 10B are extracted, a pair of a line (2) and a line (4), and a line The pair of (5) and line (7) is extracted. V2, V4, V5, and V7 are directional vectors of each line.
[0062]
In the next step 192, a specific line pair is extracted from the line pairs extracted above. In step 194, a line symmetry axis is set between the line pairs as shown in FIG. 10D, and in the next step 196, the line symmetry similarity of the line pair is determined. First, as shown in FIG. 10 (D), the line symmetry similarity is determined by determining the length of both lines along the direction parallel to the line symmetry axis by v, and the length of both lines along the direction orthogonal to the line symmetry axis. Assuming that the line interval is h, it is determined whether the aspect ratio v / h is a value within a predetermined range corresponding to the aspect ratio of the face of the person. Next, as shown in FIG. 11A, an extended area for matching is set for each line (such as a process of expanding around the line), and a curved area pattern symmetric with respect to the line symmetry axis with respect to one of the line pairs. Then, the similarity between the obtained curve area pattern and the other line expansion area pattern is determined.
[0063]
As a method of determining the degree of similarity, for example, a cumulative value of a difference absolute value or a difference square value between pixel data existing at corresponding positions of the area patterns to be matched is calculated. The similarity between areas can be determined to be large. As the data of each pixel, image data itself, binarized data, data obtained by differentiating the image data, and the like can be used.
[0064]
In step 198, assuming that the line pair is a line corresponding to both sides of the face of the person, a search is performed to determine whether a line whose direction vector is nearly vertical exists at a position corresponding to the boundary with the head. I do. For example, for the pair of the line (2) and the line (4) described above, a range 52A shown in FIG. 11B is searched as a position corresponding to the boundary with the head, and the range exists within the range 52A. The line (3) is extracted as a line whose direction vector is nearly vertical. In addition, for the pair of the lines (5) and (7), the range 52B shown in FIG. 11B is searched as a position corresponding to the boundary with the head, so that the line corresponding to the boundary with the head is searched. Is not extracted.
[0065]
In the next step 200, it is determined whether or not the line pair extracted in step 192 can be determined as a line pair corresponding to the face contour based on the above-described determination result of the line symmetry similarity. If the determination in step 200 is affirmative, the line pair is regarded as a line corresponding to both sides of the face of the person, and as shown in FIG. An elliptical face candidate area (a candidate area in which a specific portion of the extraction target graphic is presumed to be present) having a size corresponding to the line length v and the interval h between the two lines is set.
[0066]
In step 202, a positive weighting coefficient is assigned to the face candidate area set above. This weighting factor can be set so that the value increases as the line symmetry similarity increases, and increases when a line corresponding to the boundary with the head is extracted. After the process of step 202 is performed, the process proceeds to step 206. If the determination in step 200 is negative, the process proceeds to step 206 without performing any processing.
[0067]
In step 206, it is determined whether or not the above processing has been performed on all the line pairs extracted in step 190. If the determination in step 206 is negative, the process returns to step 192, and steps 192 to 206 are repeated. If the determination in step 206 is affirmative, the face candidate area setting process based on the face contour is terminated, and the process proceeds to step 108 in the flowchart of FIG.
[0068]
In step 108, face candidate area setting processing based on the face internal structure is performed. This processing will be described with reference to the flowchart of FIG. In the present embodiment, a pair of black regions estimated to correspond to a pair of eyes of a face is extracted as a shape pattern (a partial shape pattern according to claim 6) representing the internal structure of a person's face. That is, in step 210, a black region pair that can be an eye pair candidate is extracted from the black regions obtained by the above-described binarization. As shown in FIG. 12A, a pair of black regions is first extracted from each black region, as shown in FIG. 12A, in the dimensional ratio (long-short axis ratio) in the longitudinal direction (long-axis direction) and width direction (short-axis direction). ) Searches for a plurality of elliptical black regions within a predetermined range. Next, the angle in the long axis direction is obtained based on the primary moment of inertia of each of the searched black areas, and a pair of black areas in which the difference in the angle in the long axis direction is within a predetermined range is determined as a black that can be a candidate for an eye pair. Extract as a region pair.
[0069]
In step 212, a specific black region pair is extracted from the black region pair extracted above, and in the next step 214, as shown in FIG. 12B, a line perpendicular to the line connecting the centers of gravity of both black regions. An axis of symmetry is set, an extended area for matching is set around each black area, and the line symmetry similarity of the black area pair is determined in step 216. In the next step 218, it is determined whether or not the black region pair extracted in step 212 can be determined as an eye pair based on the determination result of the line symmetry similarity of the black region pair determined above.
[0070]
If the determination in step 218 is affirmative, the process proceeds to step 220, and based on the position of the black area, the size of each of the black areas, and the direction of the axis of symmetry, as shown in FIG. An area (a candidate area in which a specific portion of the extraction target graphic according to claim 6 is presumed) is set. In the next step 222, a positive weighting coefficient is assigned to the face candidate area set above. The weight coefficient can be set so that the value increases as the line symmetry similarity increases. After the process of step 222 is performed, the process proceeds to step 224. If the determination in step 218 is negative, the process proceeds to step 224 without performing any processing.
[0071]
In step 224, it is determined whether or not the above processing has been performed on all the black region pairs extracted in step 210. If the determination in step 224 is negative, the process returns to step 212, and steps 212 to 224 are repeated. If the determination in step 224 is affirmative, the face candidate area setting process based on the face internal structure ends, and the process proceeds to step 110 in the flowchart of FIG.
[0072]
In step 110, a face candidate area setting process based on the body contour is performed. This processing will be described with reference to the flowchart in FIG. In the present embodiment, it is estimated that the shape pattern (partial shape pattern according to claim 6) representing the contour of the body corresponds to a contour that is continuous from the neck to the shoulder of the person, from the shoulder to the arm, or to the lower part of the body. Extract pairs of lines. That is, in step 230, a set of lines that can be one side candidate of the body contour is extracted from the lines extracted by the face candidate area setting processing based on the face contour described above. Specifically, a pair of lines in which the distance between the end points of both lines is short and the angle at which both lines intersect within a predetermined range can be extracted as a set of lines that can be a candidate for one side of the body contour. For example, when a line as shown in FIG. 13A is extracted, as shown in FIG. 13B, lines (5) and (7) whose crossing angles are within a predetermined range (θ1 and θ2), respectively. , And the sets of lines (6) and (8) are extracted.
[0073]
In the next step 232, both lines are extended and connected to each of the set of lines extracted above, thereby generating a one-sided candidate of the body contour. Further, in step 234, for each of the one-sided candidates for the torso contour generated above, a pair of one-sided candidates for the torso contour that can be candidates for the torso contour (for example, a pair in which the concave portions of the one-sided candidates for both torso contours face each other). Is extracted.
[0074]
In step 236, a specific pair is extracted from the pair of one-sided candidates for the torso contour extracted above. In step 238, the pair of one-sided candidates for the torso contour is extracted with respect to the line symmetry axis as shown in FIG. Is set, and in the next step 240, the line symmetry similarity of the pair of one side candidates of the body contour is determined. In addition, for the lines (for example, lines (5) and (6) in FIG. 13B) estimated to correspond to the contour from the neck to the shoulder of the person, the angle (for example, in FIG. It is determined whether or not (1) in the case of the line (5) is within a predetermined range. In step 242, based on the line symmetry similarity determined above, it is determined whether a pair of one-sided candidates for the body contour can be determined as the body contour.
[0075]
If the determination in step 242 is affirmative, then in step 244, based on the position, size, interval, and direction of the line symmetry axis of both one-side candidates constituting the torso contour candidate, as shown in FIG. , An elliptical face candidate area (a candidate area in which it is estimated that the specific part of the extraction target graphic according to claim 6 is present) is set. In step 250, a positive weighting coefficient is assigned to the face candidate area set above. This weighting factor can be set so that the value increases as the line symmetry similarity determined in step 240 increases. When the process of step 240 ends, the process moves to step 248. If the determination in step 242 is negative, the process proceeds to step 248 without performing any processing.
[0076]
In step 248, it is determined whether or not the above processing has been performed on all pairs of one-sided candidates of the body contour extracted in step 234. If the determination in step 248 is negative, the process returns to step 236, and steps 236 to 248 are repeated. If the determination in step 248 is affirmative, the face candidate area setting process based on the body contour is terminated, and the process proceeds to step 112 in the flowchart of FIG.
[0077]
In step 112 of the flowchart in FIG. 2, a non-person area determination process is performed. This non-person area determination processing will be described with reference to the flowchart in FIG. In step 130, a straight line portion of the contour is detected for each of the regions divided in step 102 of the flowchart of FIG. 2, and in step 132, the ratio of the straight line portion in the contour is calculated for each region. In step 134, it is determined whether or not there is an area where the ratio is equal to or more than a predetermined value. It can be determined that a region where the ratio of the straight line portion occupying the contour is equal to or more than a predetermined value has a high probability of being a region representing an artificial object, and has a low probability of being a region corresponding to a person. Therefore, if the determination in step 134 is affirmative, in step 136 a negative weighting coefficient is assigned to a region in which the ratio of the linear portions is equal to or greater than a predetermined value, and the process proceeds to step 138.
[0078]
In step 138, a line symmetry axis extending along the vertical direction of the image is set near the center of the image in the left and right direction in each region. In step 140, the degree of line symmetry with respect to the line symmetry axis set above is set. The calculation is performed for each area. In step 142, it is determined whether or not there is an area where the degree of line symmetry is equal to or less than a predetermined value. In general, the human body is substantially line-symmetric with respect to a virtual line dividing the left and right of the human body, and it can be determined that a region where the degree of line symmetry is equal to or less than a predetermined value has a low probability of being a region corresponding to a person. Therefore, if the determination in step 142 is affirmative, a weighting coefficient with a negative sign is assigned to an area where the degree of line symmetry is equal to or less than a predetermined value in step 144, and the process proceeds to step 146.
[0079]
In step 146, the contact ratio with the outer edge of the image occupying the contour is calculated for each region. It can be determined that the area where the contact rate is equal to or more than a predetermined value is located at the periphery of the image. However, in general, the probability that a person is located substantially at the center of the image is high, Can be determined to have a low probability of being a region corresponding to a person. Therefore, if the determination in step 148 is affirmative, a weighting coefficient with a negative sign is assigned to an area where the contact ratio with the outer edge of the image is equal to or more than a predetermined value in step 150, and the process proceeds to step 152.
[0080]
In step 152, the contrast of the density inside each area (the difference between the maximum density value and the minimum density value) is calculated, and in step 154, it is determined whether or not there is an area where the density contrast is equal to or less than a predetermined value. If the contrast of the density in the area is equal to or less than a predetermined value, it is highly likely that the area is an area representing an object having a smooth or less uneven surface, and the probability of being an area corresponding to a person is low. Therefore, if the determination in step 154 is affirmative, in step 156 a negative weighting coefficient is assigned to a region where the internal density contrast is equal to or less than a predetermined value, and the process proceeds to step 158.
[0081]
In step 158, the density value of each pixel in each region is differentiated along a plurality of different directions (for example, a direction inclined ± 45 ° with respect to the horizontal direction, the vertical direction, and the horizontal direction of the image), and step 160 It is determined whether there is a region where at least one of the differential values along a plurality of directions changes regularly. The above determination is affirmed for a region where the density changes in a predetermined pattern or a region where the density repeats a predetermined change pattern. Since such a region can be determined to have a low probability of being a region corresponding to a person, if the determination in step 160 is affirmed, in step 162, the sign of the region in which the differential value changes regularly is added. Give a negative weighting factor. Thus, the non-person area determination processing ends, and the flow shifts to step 118 in the flowchart of FIG.
[0082]
In step 118, comprehensive determination of the face area is performed. That is, if the ranges set as the face candidate regions in steps 104 to 110 and the positive weighting factors assigned to each face candidate region are used, and if there is a region to which a negative weighting factor is added in step 112, , Based on the range of the area and the assigned negative weighting coefficient, a total weighting coefficient of each portion in the image is calculated. More specifically, if there is a portion in the image to which a weighting factor has been assigned by a plurality of processes, the partial range and the overall weighting factor (determined by adding or multiplying the weighting factor assigned in each process. The weighting factor corresponds to the degree of matching described in claim 6), and the image is divided into regions having the same overall weighting factor. Then, the area having the highest overall weighting coefficient is determined as the area corresponding to the face area of the person and having the highest certainty.
[0083]
In the next step 120, it is determined whether or not the processing of steps 102 to 118 has been executed a predetermined number of times. If the determination in step 120 is negative, in step 122, various control parameters are updated to change the coarse density of the image division in step 102, or the threshold for binarization in step 102 is updated. Then, the processing after step 102 is executed again. After executing Steps 102 to 118 a predetermined number of times, the determination at Step 120 is affirmed, and the routine proceeds to Step 124.
[0084]
As described above, the reason why the processing is repeated by changing the coarse density of the image division is that the area of the area corresponding to the face of the person existing in the original image is large or small (one person existing in the original image is This is because the appropriate size of each region when the original image is divided into a plurality of regions differs depending on the size of the corresponding region).
[0085]
In other words, if the area of the region corresponding to the face of a person existing in the original image is large, the original image is divided densely (the size of each region is reduced), and the region corresponding to the original face is divided finely. Since the area corresponding to each part of the person is too finely divided, the area corresponding to each part of the person is not properly separated as a whole, but if the original image is roughly divided (the size of each area is large), A divided region having a high degree of coincidence with the region corresponding to the face can be obtained, and the region corresponding to each part of the person can be appropriately separated.
[0086]
On the other hand, if the area of the area corresponding to the face of the person present in the original image is small, when the original image is roughly divided, the area corresponding to the original face is hidden by the area corresponding to the background, and Although it is not possible to properly separate the face and the area corresponding to each part, if the original image is divided densely, the area corresponding to the person's face and each part can be properly separated. Therefore, by repeatedly performing the face region extraction process a plurality of times while changing the coarse density of the image division stepwise, regardless of the size of the region corresponding to the person's face present in the original image, The corresponding area can be extracted with high accuracy.
[0087]
In addition, when changing the image division control parameter, for example, when dividing an original image by clustering using a one-dimensional or two-dimensional histogram, the step width of a threshold value when quantizing data at the time of creating a histogram is changed. When the histogram is made coarse (large), the result of dividing the original image using the histogram also becomes coarse (the size of the divided region becomes large), and when the step width of the threshold is made fine (small), the original image is formed using the histogram. Since the result of dividing the image becomes finer (the size of the divided region becomes smaller), the step width of the threshold value when quantizing is used as the image division control parameter, and the step width of the threshold value is changed. By doing so, the coarse density of image division can be changed.
[0088]
When the original image is divided by using a pixel having a large density difference or a color difference from an adjacent pixel among the pixels constituting the original image as a region boundary, it is determined that the density difference or the color difference is large. If the threshold value for the division is increased, the result of dividing the original image becomes coarse (the size of the divided region becomes large), and if the threshold value is decreased, the result of dividing the original image becomes fine (the size of the divided region becomes small). Therefore, the threshold value can be used as an image division control parameter.
[0089]
Also, in the division of the original image, regardless of how the division is made, if the resolution of the original image is made coarse, the result of dividing the original image becomes coarse, and if the resolution is made fine, the result of dividing the original image becomes coarse. Alternatively, the image resolution may be used as the image division control parameter. As an example of changing the image resolution, when reducing the image resolution, for each pixel block composed of m × n pixels, the average of the density or color of all the pixels constituting each block is calculated for each block. Can be taken as the representative value of In the method of changing the coarse density of the image division by changing the image resolution as described above, there is also an advantage that the processing time can be reduced because the number of data to be processed is reduced particularly when the image resolution is reduced. . Similar effects can be obtained by performing a smoothing process instead of changing the image resolution. In the smoothing process, as the degree of smoothing increases, the averaging of the density and the color with the surrounding pixels progresses, and the result of dividing the original image becomes coarse as in the case where the resolution is reduced.
[0090]
In step 124, based on the face regions determined in the predetermined number of processes and the weighting factors assigned to them, the region with the highest accuracy, which is finally the region corresponding to the human face region, is first determined. The determination is finally made in the same manner as in step 118. Then, as the determination result in the next step 126, the R, G, and B photometric data of the finally determined area are output to the exposure calculator 44, and the face area extraction processing ends.
[0091]
In the above-described face area extraction processing, a face candidate area estimated to correspond to a person's face is not determined based on the density or color of the area. Therefore, even if the color balance of the area changes according to the film type of the negative film 12, the type of light source, and the shooting conditions such as backlight, the result of the face area extraction processing is changed by this influence. There is no. Further, the above-described face area extraction processing is applicable even when the original image is a monochrome image. Further, in the above-described face area extraction processing, the processing of setting a face candidate area based on a shape pattern specific to each part of a person (steps 104 to 110 in the flowchart of FIG. 2) is basically performed by using a concave part and a convex part, respectively. It consists of extracting a part and determining the line symmetry similarity. Therefore, if the software and the hardware are shared by using this, the simplification of the software and the simplification of the hardware configuration can be realized.
[0092]
On the other hand, in the exposure amount calculating section 44, the R, G, and B photometric data extracted by the face area extracting section 40 as described above, and the screen average density D of one frame calculated by the average density calculating section 42._i(I = R, G, or B) and the appropriate exposure amount E according to the following equation:_iAnd outputs the result to the driver 46. The driver 46 sets the proper exposure E_i, The color correction filter 18 is moved to a position corresponding to the appropriate exposure amount.
[0093]

However, the meaning of each symbol is as follows.
[0094]
LM: Magnification slope coefficient. It is set in advance according to the enlargement magnification determined by the type of the negative film and the print size.
[0095]
CS: color slope coefficient. It is prepared for each type of negative film, and is available for underexposure and overexposure. It is determined whether the average density of the image frame to be printed is under or over the standard negative density value, and either under exposure or over exposure is selected.
[0096]
DN: standard negative density value.
D: Density value of an image frame to be printed.
[0097]
PB: Correction balance value for standard color paper. It is determined according to the type of color paper.
[0098]
LB: Correction balance value for a standard printing lens. It is determined according to the type of lens used for printing.
[0099]
MB: Correction value (master balance value) for a change in light source light amount and a change in paper development performance.
[0100]
NB: Negative balance (color balance) value determined by the characteristics of the negative film.
[0101]
K₂: Color correction amount.
K₁: A density correction amount represented by the following equation.
[0102]
(Equation 1)

[0103]
Here, Ka and Kb are constants, and FD is a face area average density.
Further, the density correction amount K in the above equation (8)₁Is the correction value obtained by the film verification device, and the color correction amount K₂May be expressed using the face area average density as follows.
[0104]
(Equation 2)

[0105]
Here, Kc is a constant.
Further, the density correction amount K in the above equation (1)₁, Color correction amount K₂Is the correction amount obtained by the film verification apparatus, and the average density D of the print frame of the equation (1) is_iIs the average density FD of the face area._iAnd the exposure amount may be obtained.
[0106]
In addition, the density (or color) of each pixel of the area determined as the face area in the face area extraction processing is weighted according to the weight coefficient to obtain a weighted average value, and the exposure is performed using the weighted average value. Quantity E_iIs calculated, the weighting coefficient given to the area determined to be the face area is calculated as the exposure amount E._iMay be reflected.
[0107]
In the above description, the negative film 12 has been described as an example of a recording medium for an original image. However, an image recorded on another film such as a positive film or various recording media such as paper can be used as the original image. . Also, color paper has been described as an example of a copying material, but other materials such as paper may be applied.
[0108]
In the above, in the face candidate area setting processing by head extraction, as a shape pattern specific to the head of a person, a convex part corresponding to the top of the person and a concave part corresponding to the boundary between the head and the face of the person are used. Although it was used, it is not limited to this. For example, in order to extract a head without hair or a head with low hair density, area division by a method other than binarization or edge detection of an image is performed, and a convex pattern is formed as a shape pattern unique to the head. The head may be extracted using only the part.
[0109]
Further, in the above description, binarization has been described as an example of a method of dividing an image into a plurality of regions. However, the present invention is not limited to this, and a histogram as described in JP-A-4-346332 is not limited thereto. May be used for division. In addition, the density value of each pixel of the image is set in a plurality of different directions (for example, as shown in FIG. 14A, in the horizontal direction (0 °), the vertical direction (90 °), and the horizontal direction of the image). The edge of the image is detected by differentiating along the (± 45 ° tilt direction), and instead of the line described in the above embodiment, the edge of the edge obtained by performing a well-known thinning process on this edge. The processing may be performed using the center line.
[0110]
The differentiation along each direction can be realized by using each of the four types of differential filters shown in FIG. 14B, and the edge of the image is detected when the magnitude of the differential filter output is a predetermined threshold value. When it is larger than the target pixel, the target pixel can be regarded as an edge. Further, the edge extraction processing may be repeatedly performed while changing the threshold value for edge determination stepwise.
[0111]
Further, in the above description, the case where the pair of eyes is used as the shape pattern representing the internal structure of the face has been described as an example. However, the present invention is not limited to this. Alternatively, the face candidate area may be set comprehensively.
[0112]
Furthermore, in the above description, as an example of a shape pattern representing a body contour, an example of extracting a pair of lines corresponding to a continuous contour from a person's neck to a shoulder, a shoulder to an arm part, or a lower body is described. Instead, a pair of lines corresponding to the contour from the neck to the shoulder of the person and a pair of lines corresponding to the contour from the shoulder to the arm of the person may be separately detected.
[0113]
In the above description, the face candidate area setting processing based on the shape pattern specific to each part of the person and the non-person area determination processing (steps 104 to 112 in the flowchart of FIG. 2) are sequentially executed. Is not affected by the processing results of the other processes, so a processing unit for performing each of the above processes is separately provided, and as preprocessing, image division into a plurality of regions or image edge detection, concave portions, convex portions May be performed in parallel by each processing unit after the extraction of As a result, the processing time can be significantly reduced.
[0114]
[Second embodiment]
Next, a second embodiment of the present invention will be described. Since the second embodiment has the same configuration as the first embodiment, the same reference numerals are given to the respective portions, and the description of the configuration is omitted, and the operation of the second embodiment will be described below.
[0115]
In the second embodiment, the state of the figure representing the person on the image (corresponding to the extraction target figure in claim 6) is classified in advance into a plurality of states. The state of the figure representing a person may be, for example, the size (or area) of the figure on an image, the orientation of the figure on an image (the vertical direction of the person represented by the figure on the image may be any direction). ), The angle of the person as the subject with respect to the visual axis at the time of photographing (in which direction the person represented by the figure is oriented along the left and right direction of the person: (The direction of the face changes from front to front, oblique, and sideways).
[0116]
For example, when classifying the state of a figure representing a person according to the size L or the area S of the figure on the image, the size L is large (L₁~ L₂) Or medium (L₂~ L₃) Or small (L₃~ L₄) Or the area S on the image is large (S₁~ S₂) Or medium (S₂~ S₃) Or small (S₃~ S₄). When the classification is performed according to the orientation of the figure on the image, the orientation can be classified, for example, in units of 90 ° or 45 °. Further, in the case of classifying according to the angle of the person represented by the figure with respect to the visual axis at the time of shooting, the classification can be made according to whether the angle is front, diagonal or horizontal. Further, the size and orientation of the figure representing the person and the angle of the person represented by the figure with respect to the visual axis at the time of photographing may be combined for classification. As a result, the state of the figure representing the person on the image is classified into a plurality of states.
[0117]
Further, in the second embodiment, the shape pattern (corresponding to the partial shape pattern of claim 6) that forms a figure representing a person, such as the outline of the head of the person, the outline of the face, the internal structure of the face, and the outline of the body, is described. For each of them, a plurality of types of search parameters that define search conditions for searching for each shape pattern are determined according to the plurality of classified states.
[0118]
As an example, when the state of a figure representing a person on an image is classified according to the size or area of the figure, a search parameter to be changed according to the size or area is image resolution (or image division). ), And the length of the vector for calculating the curvature of the contour of the black area when searching for the shape pattern representing the head contour (the length of the vector P1P0 and the vector P0P2 shown in FIG. 8C) And the allowable range of the length of the contour to be searched for as the head outline, the face outline or the torso outline, and the allowable range of the distance between candidate pairs to be calculated when calculating the line symmetry similarity.
[0119]
As described above, if the size or area of a figure representing a person is small, the resolution is increased, and if the size or area is large, the resolution is reduced (coarse). Good. Further, the length of the vector may be shortened when the size or area is small, and may be increased when the size or area is large, and the calculation of the allowable range of the contour length and the line symmetry similarity may be performed. The permissible range of the target candidate pair distance may be shortened when the size or the area is small, and increased when the size or the area is large.
[0120]
Further, when the state of a figure representing a person on an image is classified according to the direction of the figure, a search parameter to be changed according to the direction includes a line object when searching for a shape pattern representing a face contour. The direction of the axis (see FIG. 10D), the direction of the line target axis when searching for the shape pattern representing the eye pair as the internal structure of the face (see FIG. 12B), and the shape pattern representing the body contour The direction of the line target axis at the time of the search (see FIG. 13B) is exemplified. By changing the parameters according to the orientation, the number of combinations of candidate pairs as a calculation target of the line symmetry similarity is narrowed down.
[0121]
Further, when the state of the figure representing a person on the image is classified according to the angle of the person represented by the figure with respect to the visual axis at the time of shooting, as the search parameter to be changed according to the angle, the eye pair is The distance between the eye candidate regions in the search for the shape pattern to be represented (see FIG. 12B) and the like.
[0122]
Next, with reference to the flowchart of FIG. 15, a description will be given of the face area extraction processing according to the second embodiment, only the parts different from the flowchart of FIG.
[0123]
In the second embodiment, after capturing image data in step 100, in step 101, a search parameter corresponding to a predetermined state among the plurality of classified states is captured. The predetermined state is a state having the highest frequency of appearance in an image recorded on the negative film 12 (for example, the size of a figure representing a person: medium, the orientation of the figure: person Is the direction in which the vertical direction of the image coincides with the longitudinal direction of the image, and the angle of the person with respect to the visual axis at the time of photographing represented by the figure: front.
[0124]
In the next steps 102 to 112, a face candidate area is set in the same manner as in the first embodiment based on the search parameters fetched in step 101. As described above, since the search parameters used in the processing of steps 102 to 112 are search parameters corresponding to a predetermined state among a plurality of states classified in advance, the range of the shape pattern as a search target is narrowed down. Thus, the process is completed in a shorter time than in the first embodiment. In step 118, a region having a high degree of certainty, which is a region corresponding to a person's face region, is determined in the same manner as in the first embodiment. In the processing of step 118, the degree of matching is determined for each candidate area according to claim 6, and a candidate area having a high probability that a human face as a specific part of the extraction target graphic is present is determined. It corresponds to.
[0125]
In the next step 119, it is determined whether or not a region having a very high degree of certainty, which is a region corresponding to the face of a person, has been extracted by the above processing. This determination can be made based on whether or not there is a face candidate area whose overall weight coefficient (the degree of matching of claim 6) calculated by the face area comprehensive determination in step 118 is equal to or greater than a predetermined value. If the above determination is affirmed, the process proceeds to step 124, where an area corresponding to the extracted person's face and having a very high degree of certainty is determined as a face area, and a determination result is output and processed in step 126. To end.
[0126]
On the other hand, if the determination in step 119 is negative, the process proceeds to step 121, and it is determined whether the process has been performed using all search parameters corresponding to each state. If the determination is negative, the process proceeds to step 123, in which search parameters corresponding to a state where processing has not been executed and the frequency of appearance is high are fetched, and the process returns to step 102. Thereby, the processing from step 102 onward is repeated according to the newly acquired search parameters.
[0127]
As described above, since the processing is repeated using the search parameters corresponding to the states having a high frequency of appearance in order, the determination in step 191 is affirmed before the determination in step 121 is affirmed, that is, the state corresponds to each of the classified states. Before all search parameters are used, there is a high possibility that a region having a very high degree of certainty, which is a region corresponding to a human face, is extracted, and the average processing time of the face region extraction processing is reduced. If the determination in step 119 or step 121 is affirmative, the face area comprehensive determination is performed in step 124 as in the first embodiment, the determination result is output in step 126, and the process ends.
[0128]
In the above description, the distance between the eye candidate regions is used as an example of a search parameter that is changed according to the angle of the person with respect to the visual axis at the time of shooting. Then, when searching the estimated area and determining the consistency as a human face, as a search parameter to be changed according to the angle, the position of the eye candidate area in the face candidate area or the line symmetric axis The degree of positional deviation, the ratio of the distance between the pair of eye candidate regions to the width of the face candidate region, and the like may be applied, and the allowable range of each search parameter may be changed according to the angle.
[0129]
Further, in the above description, the processing is performed using a plurality of types of search parameters in order, but the present invention is not limited to this, and the processing of steps 102 to 118 in the flowchart of FIG. A plurality of processing units may be provided (see FIG. 16), and each processing unit may perform processing in parallel using each of the plurality of types of search parameters corresponding to each state. In the example of FIG. 16, the states of the figure representing the person are classified into n states 1 to n, and n processing units 60₁~ 60_nIn the flowchart of FIG. 15, the face area comprehensive determination in step 124 and the determination result output in step 126 are performed by the face area overall determination unit 62. When configured as described above, each processing unit 60₁~ 60_nCan be completed in a very short time, so that the processing time of the face region extraction processing can be significantly reduced regardless of the state of the figure representing the person on the image.
[0130]
Further, in the above description, an example in which a region corresponding to a person's face is extracted as a specific part of the extraction target graphic has been described. However, the present invention is not limited to this. (The entire corresponding area) may be extracted.
[0131]
In the above description, a figure representing a person is applied as the extraction target graphic. However, the present invention is not limited to this. For example, a figure representing an artificial object may be applied as the extraction target graphic. Artifacts existing in an image recorded on a photographic film generally have unknown shapes and the like, but the partial shape of the figure representing the artifact is determined by the non-human area determination processing (FIG. 7) as described above. It is empirically known that it is often composed of straight lines, circular arcs having a constant curvature, and the like. Therefore, if these are used as the partial shape patterns, it is possible to extract a graphic representing an artificial object as the graphic to be extracted. When an image recorded on a photographic film is printed on a copy material, for example, exposure to the copy material is performed by reducing the weight of an area determined to have a high probability that a figure representing an artifact exists. The amount can be determined.
[0132]
The image from which a graphic is extracted by applying the present invention is not limited to an image recorded on a photographic film. As an example, in mass production of parts, products, and the like, while imaging the situation in which the produced parts, products, and the like are being sequentially conveyed, an image representing the conveyed situation is extracted at a predetermined timing from the imaging signal. It is also possible to extract, from the extracted image, a graphic representing the part or product as the graphic to be extracted. In this case, since the shapes of components, products, and the like are known in advance, the partial shape pattern can be easily determined. The region where the extraction target graphic extracted according to the present invention exists can be used, for example, for automatically inspecting a produced part or product.
[0133]
Although the embodiments of the present invention have been described above, the above-described embodiments include, in addition to the technical matters described in the claims, embodiments of the following technical matters.
[0134]
(1) When the shape pattern representing the contour of the head of the person is to be searched, the contour of the area divided by the invention of claim 3 or the center of the edge obtained by the invention of claim 4 A concave portion and a convex portion are detected from the line, and the concave portion or the convex portion is formed on the head of the person based on the characteristic amount (curvature, position, size, direction) of one or more detected concave portions or convex portions. A method for determining an exposure amount, wherein the shape pattern is searched for by determining whether or not the shape pattern corresponds to the contour of the part.
[0135]
(2) When a shape pattern representing the contour of a person's face is to be searched, the contour of the region divided by the invention of claim 3 or the center line of the edge obtained by the invention of claim 4 The concave and convex portions are further detected, and the two or more concave and convex portions are detected based on the relationship between the characteristic amounts (curvature, position, size, direction) of the two or more concave portions or convex portions, and the two are determined by the degree of line symmetry. A method for determining the amount of exposure, wherein the shape pattern is searched by determining whether or not the concave portion or the convex portion corresponds to the contour of a person's face.
[0136]
(3) When a shape pattern representing an eye pair among shape patterns representing a person's face structure is to be searched, the area divided by the invention described in claim 3 or the area obtained by the invention described in claim 4 is obtained. An elliptical area is extracted from the area divided by the center line of the extracted edge, and a similarity degree equal to or greater than a predetermined value is obtained by a matching process using the feature amounts (shape and density) of each of the plurality of extracted elliptical areas. The pair of elliptical regions is detected, and it is determined whether or not the pair of elliptical regions corresponds to the pair of eyes of a person's face based on the degree of line symmetry of the detected pair of elliptical regions. A method for determining the amount of exposure, wherein
[0137]
(4) When the shape pattern representing the contour of the body of the person is to be searched, the contour of the area divided by the invention of claim 3 or the center line of the edge obtained by the invention of claim 4 The contour from the neck to the shoulder of the person is detected based on the relationship between the feature amounts (curvature, position, size, direction) of the two or more detected recesses or protrusions. , And detecting a plurality of sets of concave or convex portions corresponding to the contour from the shoulder to the arm portion or the lower part of the torso, and further determining whether or not the set corresponds to the contour of the torso of the person based on the degree of line symmetry between the sets. A search for the shape pattern.
[0138]
(5) An area divided by the invention according to the third aspect, or divided by the center line of the edge obtained by the invention according to the fourth aspect, as a low-probability area corresponding to the face of a person in the original image. Among the regions, the ratio of the straight line portion included in the outline of the region is a predetermined value or more, or the line symmetry is a predetermined value or less, or the number of irregularities is a predetermined value or more, or the contact ratio with the image outer edge is It is characterized by determining a region equal to or more than a predetermined value, a region where the contrast of the internal density is equal to or less than a predetermined value, or a region where the internal density changes in a predetermined pattern or where a predetermined change pattern is repeated. Exposure amount determination method.
[0139]
【The invention's effect】
As described above, the invention according to claim 1 searches for a shape pattern specific to a predetermined portion of a person existing in an original image, and detects the size and direction of the detected shape pattern and the position of the detected shape pattern. An area estimated to correspond to the face of a person is set according to the relationship, and a weighting factor is assigned to the set area, with each of the shape patterns unique to each part of the person being searched, and the face of the person is set. The range of each of the plurality of regions set as the region estimated to correspond to, and the region having the highest degree of certainty corresponding to the face of the person in the original image is determined based on the weighting factors assigned to the plurality of regions, Since the exposure amount is determined based on at least one of the color and the density of the determined area, the area corresponding to the person's face in the original image and the area corresponding to the person's face are not affected by the colors of the surrounding areas. Extracting only an area, the exposure amount can be determined so properly baked region, has an excellent effect that.
[0140]
According to a third aspect of the present invention, the original image is divided into a plurality of regions composed of a plurality of pixels having the same or similar density or color, and a shape pattern is searched using the outline of each of the plurality of regions. Therefore, there is an effect that a shape pattern specific to each part of the person can be efficiently detected.
[0141]
According to the fourth aspect of the present invention, an edge present in the original image is detected, a center line of the detected edge is obtained, and a shape pattern is searched using the obtained center line. Can be detected efficiently.
[0142]
The invention according to claim 5 assigns a positive weighting coefficient to an area estimated to correspond to a person's face, and determines and determines an area with low accuracy corresponding to the person's face in the original image. Since the area is given a negative weighting factor with a sign, the probability that a region corresponding to a person's face is extracted as a region having the highest accuracy corresponding to a person's face is further improved.
[0143]
In the invention according to claim 6, a plurality of partial shape patterns are determined in advance based on a plurality of characteristic partial shapes constituting the extraction target graphic, the partial shape patterns are searched, and the size of the detected partial shape pattern is determined. Setting a candidate region in which it is assumed that a specific portion or the whole of the extraction target graphic in the image is present according to the position of the partial shape represented by the partial shape pattern in the extraction target graphic. This is performed for each of the patterns, and based on the matching degree obtained for each of the candidate regions, a candidate region having a high probability that the specific portion or the entirety of the extraction target graphic exists is determined. This has an excellent effect that the target graphic can be properly extracted without being affected by the colors of the region where the extraction target graphic exists and the surrounding region.
[0144]
According to a seventh aspect of the present invention, a plurality of types of parameters are determined in advance as parameters defining search conditions for each of the plurality of partial shape patterns in accordance with the state of the extraction target graphic on the image. Is performed for each of the plurality of partial shape patterns, and a process of determining and extracting a candidate region having a high probability of the existence of the extraction target graphic based on the degree of matching is performed by a plurality of processes. A candidate that is determined to have a high degree of certainty that the extraction target graphic exists, as a result of performing each of the plurality of types of parameters corresponding to the state, or performing the processing using the parameters corresponding to the predetermined state. If there is no area, the above processing is repeated using parameters corresponding to a state different from the predetermined state, and it is determined that the probability that the extraction target graphic exists is high. Since it was to terminate the process if the candidate region there, in addition to the above effects, it is possible to shorten the processing time, an effect that.
According to the ninth aspect of the present invention, a shape pattern specific to a predetermined portion of a person existing in an original image is searched for, and the size and direction of the detected shape pattern and the positional relationship between the face of the person and the predetermined portion are detected. A region that is estimated to correspond to the face of a person is set according to the above, and a weighting coefficient is assigned to the set region, as a search target for a plurality of types of shape patterns specific to each part of the person. Based on the respective ranges of the plurality of regions set as the regions estimated to correspond to the face and the weight coefficient assigned to each of the plurality of regions, the region having the highest likelihood corresponding to the face of the person in the original image is determined. Since the determination is made, an excellent effect that the extraction accuracy of the region corresponding to the person's face can be improved without being affected by the colors of the region corresponding to the person's face in the original image and the surrounding regions. Having
[Brief description of the drawings]
FIG. 1 is a schematic configuration diagram of a photographic printing apparatus according to an embodiment.
FIG. 2 is a flowchart illustrating a main routine of a face area extraction process executed by a face area extraction unit according to the first embodiment.
FIG. 3 is a flowchart illustrating a face candidate area setting process based on head extraction.
FIG. 4 is a flowchart illustrating a face candidate area setting process based on a face contour.
FIG. 5 is a flowchart illustrating a face candidate area setting process based on a face internal structure.
FIG. 6 is a flowchart illustrating a face candidate area setting process based on a body contour.
FIG. 7 is a flowchart illustrating a non-person area determination process.
8A and 8B show details of face candidate area setting processing by head extraction, wherein FIG. 8A shows an original image, FIG. 8B shows a trace of a contour of a black area extracted by binarization, and FIG. (D) is an image diagram for explaining extraction of a concave portion and a convex portion.
FIG. 9 shows details of face candidate area setting processing by head extraction, in which (A) calculates the feature amounts of the concave and convex parts, (B) determines the head, and (C) sets the face candidate area. It is an image figure for each explaining.
10A and 10B show details of a face candidate area setting process based on a face contour; FIG. 10A shows an original image, FIG. 10B shows extraction and division of a concave part and a convex part, and FIG. (D) is an image diagram for explaining each determination of the line symmetry similarity.
FIGS. 11A and 11B show details of face candidate area setting processing based on face contours, in which (A) shows determination of line symmetry similarity, (B) shows a search for a boundary with the head, and (C) shows setting of a face candidate area. It is an image figure for each explaining.
FIGS. 12A and 12B show details of face candidate region setting processing based on a face structure, in which (A) extracts a black region that can be a candidate for an eye, (B) determines the line-symmetric similarity of a black region pair, and (C) shows It is an image figure for each explaining setting of a face candidate field.
FIGS. 13A and 13B show details of face candidate region setting processing based on a torso contour, wherein FIG. 13A shows a line extracted in the pre-processing, FIG. 13B shows determination of line symmetry similarity, and FIG. It is an image figure for each explaining.
FIG. 14A is a conceptual diagram illustrating an example of a differential direction in edge detection, and FIG. 14B is a conceptual diagram illustrating an example of a differential filter for performing differentiation in each direction.
FIG. 15 is a flowchart illustrating a face area extraction process according to the second embodiment.
FIG. 16 is a schematic block diagram illustrating an example of a configuration of a face area extraction unit for executing face area extraction processing in parallel.
[Explanation of symbols]
10. Photo printing equipment
12 Negative film
18. Color correction filter
28 color paper
32 CCD image sensor
40 face area extraction unit
44 Exposure calculator

Claims (9)

A shape pattern specific to a predetermined portion of the person present in the original image is searched, and the size and orientation of the detected shape pattern and the position of the person and the predetermined position correspond to the face of the person according to the positional relationship between the predetermined portion. Then, an area to be estimated is set, and a weighting coefficient is given to the set area, and a plurality of types of shape patterns unique to each part of the person are respectively performed as search targets,
Based on the respective ranges of the plurality of regions set as the regions estimated to correspond to the person's face, and the weighting factor assigned to each of the plurality of regions, the likelihood corresponding to the person's face in the original image is the highest. Determine the area,
Determine the amount of exposure to the copy material based on at least one of the color or density of the determined area,
Exposure determination method. The shape pattern unique to each part of the person,
Including a shape pattern representing a contour of a person's head, a shape pattern representing a contour of a person's face, a shape pattern representing an internal structure of a person's face, and a shape pattern representing a contour of a person's torso,
2. The method according to claim 1, wherein the exposure amount is determined. The search for the shape pattern is performed by dividing the original image into a plurality of regions composed of a plurality of pixels having the same or similar density or color, and using the outline of each of the plurality of regions.
2. The method according to claim 1, wherein the exposure amount is determined. The search for the shape pattern is performed by detecting an edge present in the original image, obtaining a center line of the detected edge, and using the obtained center line.
2. The method according to claim 1, wherein the exposure amount is determined. A sign is assigned a positive weighting coefficient to an area estimated to correspond to the set human face,
A region having a low degree of accuracy corresponding to the face of a person in the original image is determined, and a negative weighting coefficient is assigned to the determined region,
2. The method according to claim 1, wherein the exposure amount is determined. Predetermining a plurality of partial shape patterns based on a plurality of characteristic partial shapes constituting the extraction target graphic,
A partial shape pattern existing in the image is searched for, and the size and direction of the detected partial shape pattern and the extraction target graphic in the image are specified according to the position of the partial shape represented by the partial shape pattern in the extraction target graphic. By setting a candidate region that is estimated to exist part or whole, for each of the plurality of partial shape patterns,
For each set candidate area, determine the degree of consistency as a region where a specific portion or the entirety of the extraction target graphic exists based on the relationship with the detected each partial shape pattern,
Based on the matching degree obtained for each candidate area, determine a candidate area having a high probability that a specific portion or the whole of the extraction target graphic exists,
A graphic extraction method for extracting a candidate area determined to have a high probability that a specific part or the whole of an extraction target graphic exists. While preliminarily classifying the state of the extraction target graphic on the image into a plurality of states,
For each of the plurality of partial shape patterns, as a parameter that defines a search condition when searching for a partial shape pattern, a plurality of types of parameters are determined in advance according to the plurality of classified states,
Searching for a partial shape pattern present in an image, setting the candidate area is performed for each of the plurality of partial shape patterns, obtaining a degree of matching for each of the set candidate areas, and specifying a specific portion or the entirety of the extraction target graphic. The process of determining and extracting a candidate region having a high probability of existence is performed using each of a plurality of types of parameters corresponding to the plurality of states,
Alternatively, as a result of performing the processing using a parameter corresponding to a predetermined state among the plurality of states, there is no candidate area determined to have a high probability that the specific part or the whole of the extraction target graphic exists. In such a case, there was a candidate area that was determined to have a high probability that the specific part or the whole of the extraction target graphic was present with high repetition of performing the processing using parameters corresponding to a state different from the predetermined state. 7. The graphic extraction method according to claim 6, wherein said processing is terminated in said case. The state of the extraction target graphic on the image is
8. The method according to claim 7, further comprising at least one of a size of the extraction target graphic on the image, an orientation of the extraction target graphic on the image, and an orientation of an object represented by the extraction target graphic. . A shape pattern specific to a predetermined part of the person present in the original image is searched for, and corresponds to a person's face according to the size and orientation of the detected shape pattern and the positional relationship between the person's face and the predetermined part. Then, an area to be estimated is set, and a weighting coefficient is given to the set area, and a plurality of types of shape patterns specific to each part of the person are performed as search targets, respectively.
Based on the respective ranges of the plurality of regions set as the regions estimated to correspond to the person's face and the weighting factors respectively assigned to the plurality of regions, the probability that the person's face in the original image corresponds to the highest is highest. Judge the area
Face area determination method.

Images